JP4513219B2 - Prefix nonlinear distortion compensator - Google Patents

Description

ｘ（ｔ）、ｙ（ｔ）およびｚ（ｔ）は、それぞれ帯域信号Ｓ_IN、Ｓ₀およびＳ_OUTの複素包絡線 (ベースバンド信号)である。ｊは虚数を表わす。ｆ_RFは無線周波数（Radio Frequency）を表わす。ｔは時刻を表わす。
【００２５】
［非線形歪補償器の動作原理］
Ｐ＝｜ｙ（ｔ）｜²／２として、増幅器１６のベースバンド領域における入出力特性をｆを複素関数としてｆ（Ｐ）と置くと、ベースバンド信号ｙ（ｔ）およびｚ（ｔ）の関係は、次式（４）のようにあらわすことができる。
【００２６】
一般に、非線形歪補償器では、ｆ（Ｐ）に対し、次式（５）の関係を満たす入出力特性ｇ（Ｐ）を求めることを目的とする。仮に、入出力特性ｇ（Ｐ）を得ることができれば、次式（６）および（７）が成立ち、Ｓ_INに対してＳ_OUTを線形化することができる。
【００２７】
【数１】

【００２８】
本実施の形態では、ｇ（Ｐ）をＳＶＭ２６および係数更新部３２にて推定する。具体的には、ｙ（ｔ）とｚ（ｔ）より推定したｙ′（ｔ）との誤差ｅ（ｔ）が最小になるようにＳＶＭ２６の関数の係数を更新する。この計算が収束すれば、ＳＶＭ２６は増幅器１６の逆特性を有することとなる。このため、ＳＶＭ８にＳＶＭ２６と同じ係数を与えることにより、増幅器１６の非線形性を補償することができる。
【００２９】
［逆関数の計算］
本実施の形態では、増幅器１６の入出力特性の逆特性ｇ（Ｐ）をサポートベクトルマシンを用いて推定する。サポートベクトルマシン自体はこれまで主にパターン認識の分野など分類問題に適用する事例について広く研究されており、周知の技術である。しかし、本発明のようにサポートベクトルマシンを回帰問題に適用している事例はない。
【００３０】
計算のための条件を以下に整理しておく。
・非線形の歪補償のための計算は時間連続信号ではなく離散信号に対して行なわれる。ハードウェア化の際は、Ａ／Ｄ変換器を使用し、連続信号を離散信号に変換し、計算後、離散信号を連続信号にＡ／Ｄ変換器により復元する。
・信号処理は帯域系ではなくベースバンド系または中間周波数信号にて行なわれる。ベースバンド系の場合には、増幅器１６は等価的にベースバンド信号ｙ（ｔ）を受け、ｚ（ｔ）を出力する系と見なすことができる。
【００３１】
［ＳＶＭ２６の動作説明］
時刻ｋＴ（Ｔ：Ａ／Ｄ変換およびＤ／Ａ変換のサンプリング間隔）のときのＳＶＭ２６の出力ｙ′_s（ｋＴ）は、次式（８）により得られる。
【００３２】
【数２】

【００３３】
ここで、ｍは信号ベクトルおよび後述するサポートベクトルのサイズ、ｎは学習に用いたデータベクトルの数、ｐ_iはｉ番目の学習データベクトルを取得した時間である。学習データの取得タイミングｐ_iはシステムのパフォーマンスより実験的に決定する。ｚ_i（ｉ＝１，．．．，ｎ）は増幅器１６の入出力特性を学習するためのデータベクトルである。α_iはｚ_iに対応した重み係数（実数）である。一般に、α_iが零の時、対応するｚ_iは計算に加える必要がなく廃却しても構わない。逆に、α_iが非零の時、対応するｚ_iはサポートベクトルと呼ばれる。また、非零のα_iの個数をサポートベクトルマシンの次元と呼ぶ。例えば、非零のα_iの個数が７３であればサポートベクトルマシンの次元数は７３である。さらに、この次元数は学習データを増やしても一定値以上に増えることがない。この上界値のことをＶＣ（Vapnik and Chervonenkis）次元と呼ぶ。
【００３４】
Ｋ（Ｖ＿ｘ，Ｖ＿ｙ）は内積カーネルと呼ばれる関数であり、本実施の形態では、内積カーネルの線形和である式（８）を、適切なα_iを選択することにより、増幅器１６の入出力特性の逆特性ｇ（Ｐ）に適合させることを目標とする。変数の接頭辞”Ｖ＿”は、その変数がベクトルであることを表わす。このように未知の関数にシステムの入出力関数を適合させるような問題全般を総称して「回帰問題」という。次式（１２）〜（１４）は、一般的に用いられる内積カーネルの例である。
【００３５】
【数３】

【００３６】
どの内積カーネルが逆関数を表現するのに最も適しているかはシステムにより異なり、実際には動作前に何度か試験をした上で最も特性のよいものを選択する。もちろん、上記以外の関数でも逆関数ｇ（Ｐ）をよりよく推定でき、かつ期待する効果が得られるのであれば、その関数を使用してもよい。
【００３７】
また、式（８）において、複素数の実数部のみを抽出して、ＳＶＭ２６の出力の計算が行なわれる。これは、サポートベクトルマシンが複素数を取り扱うことができないという問題を解消するためである。
【００３８】
［係数更新部３２の動作説明］
係数更新部３２は、Ｒｅ｛ｙ（ｔ）｝とＲｅ｛ｙ′（ｔ）｝との誤差ｅ（ｔ）が最小になるようにα_iを制御する。α_iは次式（１５）および（１６）で表わされる最適化問題を解くことにより得られる。
【００３９】
【数４】

【００４０】
ただし、正の値Ｃは制約を破ったサンプルに対するペナルティに対応する定数である。εは雑音に対するマージンである。通常、両定数は実験的に求められる。
【００４１】
増幅器１６の入出力特性は周囲の環境に応じて変動すると考えられる。このため、入出力特性の変動に追従できるよう適応的にα_iの値を更新する必要がある。
【００４２】
係数更新部３２は、以下のようにしてα_iの値を更新する。
（１） 係数更新部３２は、ｎ−１組のデータＶ＿ｚ_iおよび係数α_iを入手しているものとする。
（２） 係数更新部３２は、ｎ個目のデータＶ＿ｚ_nを入手する。
（３） α₁，．．．，α_n-1に関しては係数を固定し、α_nのみについて式（１５）で表わされる最適化問題を解く。
（４） α_nが加わったことによる最適解からのずれ（摂動）を補正するため、α_i，（ｉ＝１，．．．，ｎ）に関して誤差分（摂動分）Δα_iを計算する。
【００４３】
上述のように新たにデータが加わったことによる係数の摂動分のみを計算することにより、すべての係数を計算し直す場合に比べ、計算量を削減することができる。
【００４４】
係数の更新方法の具体例を以下に示す。ただし、この具体例は実現の一例であって、必ずしもこの通りである必要はない。また以下では、表記を簡単にするためＴ＝１としている。このようにしても一般性を失うことはない。
【００４５】
図２を参照して、係数更新部３２は、ｎ＝１，α_n＝１，ｙ′（ｐ_n）＝０とおく（Ｓ２）。学習データＶ＿ｚ_nとｙ_s（ｐ_n）とを入手する（Ｓ４）。ただし、Ｖ＿ｚ_nおよびｙ_s（ｐ_n）は以下のように表わされる。ここで、ｄは増幅器１６による遅延時間を表わす。
【００４６】
【数５】

【００４７】
減算器３０は、推定誤差ｅ（ｐ_n）を次式（１９）に従い計算する（Ｓ６）。
ｅ（ｐ_n）＝Ｒｅ｛ｙ_s（ｐ_n）｝−Ｒｅ｛ｙ′_s（ｐ_n）｝ …（１９）
係数更新部３２は、ｅ（ｐ_n）と雑音に対するマージンε（式（１５）を参照）とを比較する（Ｓ８）。ｅ（ｐ_n）≦εであれば（Ｓ８でＹＥＳ）、α_nに０を代入し、ｎを１つインクリメントする（Ｓ１０）。その後、Ｓ４に戻る。
【００４８】
ｅ（ｐ_n）＞εであれば（Ｓ８でＮＯ）、係数α_nを次式（２０）にしたがって計算する（Ｓ１２）。
【００４９】
【数６】

【００５０】
次に、係数更新部３２は、次式（２１）に従い、係数α_i，ｉ＝１，．．．，ｎの摂動分を計算する（Ｓ１４）。ただし、α_i＝０である係数については、摂動分の計算は行なわない。
【００５１】
Δα_i＝η_i｛ｅ_i−εｓｉｇｎ（α_i）｝ …（２１）
ｅ_iは前回計算した値を以下のルール（１）〜（２）により更新して用いる。
【００５２】
（１）ｅ_iの初期値はｙ_s（ｐ_i）とする。

また、η_iは、以下の式（２３）に従い求められる。係数更新部３２は、以下の式（２４）を計算する（Ｓ１６）。
【００５３】
【数７】

【００５４】
式（２４）の計算の結果、Ｗ_n＞Ｗ_n-1であれば（Ｓ１８でＹＥＳ）、係数α_iに摂動分Δα_iを加算し、係数α_iを更新する（Ｓ２０）。Ｗ_n≦Ｗ_n-1であれば（Ｓ１８でＮＯ）、α_nに０を代入し、Ｗ_nにＷ_n-1を代入する（Ｓ２２）。Ｓ２０またはＳ２２の処理の後、係数更新部３２は、ｎを１つインクリメントし、Ｓ４に処理の制御を戻す。
【００５５】
［ＳＶＭ８の動作説明］
ＳＶＭ８は、係数更新部３２で更新された係数α_iを用いて、次式（２５）および（２６）に従い、ベースバンド信号ｘ（ｔ）をベースバンド信号ｙ（ｔ）に変換する。なお、Ｉｍ（Ｖ＿ｘ）は、ベクトルＶ＿ｘの複素数表示の虚数部を表わす。
【００５６】
【数８】

【００５７】
上述の例では、入出力特性が不明な、非線形の入出力特性を有する増幅器１６の出力の補正に対して、サポートベクトルマシンを用いたが、本発明は増幅器１６に対してのみ適用可能なわけではない。一般的に、入出力特性が不明な信号の変換回路に対しても本発明を適用することが可能である。
【００５８】
以上説明したように、本実施の形態によると、ＳＶＭ２６を用いて、増幅器１６の逆特性を有する信号を推定する。その信号とＳＶＭ８より出力される信号とに基づいて、係数更新部３２において、増幅器１６の逆特性を有する関数の係数が更新される。この構成では、ルックアップテーブルを必要としないため、ノイズフロアを低く押さえることができ、簡単な構成で前置型非線形歪補償器を提供することができる。また、サポートベクトルマシンは、ｐｔｈ−ｏｒｄｅｒ−ｐｒｅｄｉｓｔｏｒｔｅｒに比べて、計算が簡単で収束時間が早い。このため、高速処理が可能な前置型非線形歪補償器を提供することができる。
【００５９】
また、広帯域信号が入力された場合、増幅器の非線形性が信号帯域に比較して狭帯域の場合にもデータベクトルのサイズｍを適切に設定することにより補償することが可能である。
【００６０】
［実施の形態２］
図３を参照して、本発明の実施の形態２に係る前置型非線形歪補償器は、図１を参照して説明した実施の形態１に係る前置型非線形歪補償器において、ｍ・ｄ決定部４４を付加し、係数更新部３２の代わりに係数更新部４２を用いたものである。その他の構成部品は、実施の形態１と同様である。このため、それらの説明はここでは繰り返さない。
【００６１】
ｍ・ｄ決定部４４は、直交検波部６および２２に接続されている。係数更新部は、ＳＶＭ８、実数部抽出部２８、減算器３０、ＳＶＭ２６、実数部抽出器２４およびｍ・ｄ決定部４４に接続されている。
【００６２】
ｍ・ｄ決定部４４は、次式（２７）で示される評価関数ｅ_r（ｋ）が最小となるような、信号ベクトルおよびサポートベクトルのサイズｍと増幅器１６による遅延時間ｄとを求める。係数更新部４２は、ｍ，ｄ決定部４４で決定されたｍおよびｄを用いて、係数α_nの更新を行なう。係数更新部４２の動作は、実施の形態１で示したものと同様である。このため、その詳細な説明はここでは繰返さない。
【００６３】
ｅ_r（ｋ）＝｜ｚ（ｔ）−ｋｘ（ｔ）｜² …（２７）
ＳＶＭ８は、次式（２８）に従って、ベースバンド信号ｙ（ｋＴ）を求める。
【００６４】
【数９】

【００６５】
［実施の形態３］
図４を参照して、本発明の実施の形態３に係る前置型非線形歪補償器は、無線周波数信号Ｓ_INを中間周波数信号に変換するダウンコンバータ２と、ダウンコンバータ２に接続され、ダウンコンバータ２の出力する中間周波数信号をサンプリングしてデジタル信号に変換するＡ／Ｄ変換器４と、Ａ／Ｄ変換器４に接続され、サンプリングされた中間周波数信号をベースバンド信号ｘ（ｔ）に変換する直交検波器６と、直交検波器６に接続され、ベースバンド信号ｘ（ｔ）の実数部Ｒｅ｛ｘ（ｔ）｝を抽出する実数部抽出器５２と、直交検波器６に接続され、ベースバンド信号ｘ（ｔ）の虚数部Ｉｍ｛ｘ（ｔ）｝を抽出する虚数部抽出器５６と、虚数部抽出器５２に接続され、Ｒｅ｛ｘ（ｔ）｝を、後述する増幅器１６の逆特性を用いて、ベースバンド信号ｙ（ｔ）の実数部Ｒｅ｛ｙ（ｔ）｝に変換するＳＶＭ５４と、虚数部抽出器５６に接続され、Ｉｍ｛ｘ（ｔ）｝を、増幅器１６の逆特性を用いて、ベースバンド信号ｙ（ｔ）の虚数部Ｉｍ｛ｙ（ｔ）｝に変換するＳＶＭ５８と、ＳＶＭ５４および５８に接続され、変換後のベースバンド信号ｙ（ｔ）を中間周波数信号に変換する直交変調器１０と、直交変調器１０に接続され、直交変調器１０の出力する中間周波数信号をアナログ信号に変換するＤ／Ａ変換器１２とを含む。
【００６６】
前置型非線形歪補償器は、さらに、Ｄ／Ａ変換器１２に接続され、アナログの中間周波数信号を無線周波数信号Ｓ₀に変換し、増幅器１６に供給するアップコンバータ１４と、増幅器１６に接続され、増幅器１６より出力される出力信号Ｓ_OUTを中間周波数信号に変換するダウンコンバータ１８と、ダウンコンバータ１８に接続され、中間周波数信号をデジタル信号に変換するＡ／Ｄ変換器２０と、Ａ／Ｄ変換器２０に接続され、デジタル信号をベースバンド信号ｚ（ｔ）に変換する直交検波器２２と、直交検波器２２に接続され、ベースバンド信号ｚ（ｔ）の実数部Ｒｅ｛ｚ（ｔ）｝を抽出する実数部抽出器２４と、直交検波器２２に接続され、ベースバンド信号ｚ（ｔ）の虚数部Ｉｍ｛ｚ（ｔ）｝を抽出する虚数部抽出器６０とを含む。
【００６７】
前置型非線形歪補償器は、さらに、実数部抽出器２４および後述する係数更新部４２に接続され、実数部抽出器２４の出力する実数部Ｒｅ｛ｚ（ｔ）｝と、学習に用いたベースバンド信号ベクトルｚｉ（ｉ＝１，．．．，ｎ）とに基づいて、ＳＶＭ５４より出力されるベースバンド信号の実数部の推定値Ｒｅ｛ｙ′（ｔ）｝を出力するＳＶＭ２６と、虚数部抽出器６０および後述する係数更新部６６に接続され、虚数部抽出器６０の出力する虚数部Ｉｍ｛ｚ（ｔ）｝と、ベースバンド信号ベクトルｚｉ（ｉ＝１，．．．，ｎ）とに基づいて、ＳＶＭ５８より出力されるベースバンド信号の虚数部の推定値Ｉｍ｛ｙ′（ｔ）｝を出力するＳＶＭ６２と、ＳＶＭ５４およびＳＶＭ２６に接続され、ベースバンド信号の実数部の推定値Ｒｅ｛ｙ′（ｔ）｝と、ベースバンド信号ｙ（ｔ）の実数部Ｒｅ｛ｙ（ｔ）｝との差分を計算する減算器３０と、ＳＶＭ５８およびＳＶＭ６２に接続され、ベースバンド信号の虚数部の推定値Ｉｍ｛ｙ′（ｔ）｝と、ベースバンド信号ｙ（ｔ）の虚数部Ｉｍ｛ｙ（ｔ）｝との差分を計算する減算器６４と、直交検波部６および２２に接続され、信号ベクトルおよびサポートベクトルのサイズｍと増幅器１６による遅延時間ｄとを求めるｍ・ｄ決定部４４と、減算器３０、ＳＶＭ５４、ＳＶＭ２６、実数部抽出器２４およびｍ・ｄ決定部４４に接続され、減算器３０の出力、ＳＶＭ２６の出力、ＳＶＭ５４の出力、実数部抽出器２４の出力およびｍ・ｄ決定部４４の出力に基づいて、ＳＶＭ５４における、増幅器１６の逆特性を有する関数の係数を更新する係数更新部４２と、減算器６４、ＳＶＭ５８、ＳＶＭ６２、実数部抽出器６０およびｍ・ｄ決定部４４に接続され、減算器６４の出力、ＳＶＭ６２の出力、ＳＶＭ５８の出力、実数部抽出器６０の出力およびｍ・ｄ決定部４４の出力に基づいて、ＳＶＭ５８における、増幅器１６の逆特性を有する関数の係数を更新する係数更新部６６とを含む。
【００６８】
ＳＶＭ５４およびＳＶＭ５８は、ＳＶＭ８の演算のうち、実数部および虚数部についての演算のみをそれぞれ行なう。ＳＶＭ８の演算は、実施の形態１で説明したのと同様である。このため、その詳細な説明はここでは繰返さない。
【００６９】
ＳＶＭ６２は、ＳＶＭ２６でベースバンド信号の実数部の推定値Ｒｅ｛ｙ′（ｔ）｝のと同様の方法に従い、ベースバンド信号の虚数部の推定値Ｉｍ｛ｙ′（ｔ）｝を求める。ＳＶＭ２６における処理は実施の形態１で説明した通りである。
【００７０】
係数更新部６６は、係数更新部３２がＲｅ｛ｙ（ｔ）｝とＲｅ｛ｙ′（ｔ）｝との誤差ｅ（ｔ）が最小になるようにα_iを制御するのと同様に、Ｉｍ｛ｙ（ｔ）｝とＩｍ｛ｙ′（ｔ）｝との誤差が最小になるようにα_iを制御する。係数更新部３２における処理は実施の形態１で説明した通りである。
【００７１】
上述のように、本実施の形態の前置型非線形歪補償器によると、入力信号の実数部分の補正については増幅器ベースバンド出力の実数部分を、虚数部分の補正については増幅器ベースバンド出力の虚数部分を、それぞれ用いて別々のサポートベクトルマシンによって逆関数を推定し、入力信号の実数部分と虚数部分とを別々に補正する。このような構成にすることにより、補償精度が向上する。
【００７２】
ＳＶＭ５４は、次式（２９）に従って、ベースバンド信号ｙ（ｋＴ）の実数部を求め、ＳＶＭ５８は、次式（３０）に従って、ベースバンド信号ｙ（ｋＴ）の虚数部を求める。
【００７３】
【数１０】

【００７４】
［実施の形態４］
図５を参照して、本発明の実施の形態４に係る前置型非線形歪補償器は、無線周波数信号Ｓ_INを中間周波数信号に変換するダウンコンバータ２と、ダウンコンバータ２に接続され、ダウンコンバータ２の出力する中間周波数信号をサンプリングしてデジタル信号に変換するＡ／Ｄ変換器４と、Ａ／Ｄ変換器４に接続され、サンプリングされた中間周波数信号をベースバンド信号ｘ（ｔ）に変換する直交検波器６と、直交検波器６に接続され、ベースバンド信号ｘ（ｔ）を極座標表現した際の振幅Ａ_x（ｔ）を出力する振幅計算部９２と、直交検波器６に接続され、ベースバンド信号ｘ（ｔ）を極座標表現した際の位相φ_x（ｔ）を出力する位相計算部９８と、振幅計算部９２に接続され、後述する増幅器１６の逆特性を用いて、ベースバンド信号ｘ（ｔ）をベースバンド信号ｙ（ｔ）に変換するした際に、ベースバンド信号ｘ（ｔ）を極座標表現した際の振幅Ａ_x（ｔ）からベースバンド信号ｙ（ｔ）を極座標表現した際の振幅Ａ_y（ｔ）を導出するＳＶＭ９４と、振幅計算部９２に接続され、増幅器１６の逆特性を用いて、ベースバンド信号ｘ（ｔ）をベースバンド信号ｙ（ｔ）に変換するした際に、ベースバンド信号ｙ（ｔ）を極座標表現した際の位相φ_y（ｔ）からベースバンド信号ｘ（ｔ）を極座標表現した際の位相φ_x（ｔ）を減じた値Δφ_xを導出するＳＶＭ９６と、ＳＶＭ９６および位相計算部９８に接続され、ＳＶＭ９６の出力Δφ_xおよび位相計算部９８の出力φ_x（ｔ）を入力として受け、位相φ_y（ｔ）を出力する加算器１０２と、ＳＶＭ９４および加算器１０２に接続され、振幅Ａ_y（ｔ）および位相φ_y（ｔ）より複素ベースバンド信号ｙ（ｔ）を生成する複素ベースバンド信号化部１００と、複素ベースバンド信号化部１００に接続され、ベースバンド信号ｙ（ｔ）を中間周波数信号に変換する直交変調器１０と、直交変調器１０に接続され、直交変調器１０の出力する中間周波数信号をアナログ信号に変換するＤ／Ａ変換器１２とを含む。
【００７５】
前置型非線形歪補償器は、さらに、Ｄ／Ａ変換器１２に接続され、アナログの中間周波数信号を無線周波数信号Ｓ₀に変換するアップコンバータ１４と、アップコンバータ１４に接続され、無線周波数信号Ｓ₀を増幅して、出力信号Ｓ_OUTを得る増幅器１６と、増幅器１６に接続され、出力信号Ｓ_OUTを中間周波数信号に変換するダウンコンバータ１８と、ダウンコンバータ１８に接続され、中間周波数信号をデジタル信号に変換するＡ／Ｄ変換器２０と、Ａ／Ｄ変換器２０に接続され、デジタル信号をベースバンド信号ｚ（ｔ）に変換する直交検波器２２と、直交検波器２２に接続され、ベースバンド信号ｚ（ｔ）を極座標表現した際の振幅Ａ_z（ｔ）を計算する振幅計算部１１８と、直交検波器２２に接続され、ベースバンド信号ｚ（ｔ）を極座標表現した際の位相φ_z（ｔ）を計算する位相計算部１２０とを含む。
【００７６】
前置型非線形歪補償器は、さらに、振幅計算部１１８に接続され、振幅Ａ_z（ｔ）から、ベースバンド信号ｙ（ｔ）を極座標表現した際の振幅Ａ_y（ｔ）の推定値Ａ′_y（ｔ）を計算するＳＶＭ１１２と、振幅計算部１１８に接続され、振幅Ａ_z（ｔ）から、ベースバンド信号ｙ（ｔ）を極座標表現した際の位相φ_y（ｔ）の推定値φ′_y（ｔ）から位相φ_z（ｔ）を減算した値Δφ_zを計算するＳＶＭ１１４と、ＳＶＭ１１２に接続され、振幅Ａ_y（ｔ）から振幅Ａ′_y（ｔ）を減算した値を求める減算器１０６と、ＳＶＭ１１４および位相計算部１２０に接続され、位相φ_z（ｔ）と値Δφ_z（ｔ）とを加算して、位相φ′_y（ｔ）を求める加算器１１６と、加算器１０２および加算器１１６に接続され、位相φ_y（ｔ）から位相φ′_y（ｔ）を減じた値を求める減算器１１０と、直交検波器６に接続され、信号ベクトルおよびサポートベクトルのサイズｍと増幅器１６による遅延時間ｄとを求めるｍ・ｄ決定部４４と、ｍ・ｄ決定部４４、減算器１０６および振幅計算部１１８に接続され、ｍ・ｄ決定部４４の出力、減算器１０６の出力および振幅Ａ_z（ｔ）に基づいて、ＳＶＭ９４の係数を更新する係数更新部１０４と、ｍ・ｄ決定部４４、減算器１１０および位相計算部１２０に接続され、ｍ・ｄ決定部４４の出力、減算器１１０の出力および位相φ_z（ｔ）に基づいて、ＳＶＭ９６の係数を更新する係数更新部１０８とを含む。
【００７７】
実施の形態１と同一の構成部品には同一の参照符号を付している。その動作も同一であるため、ここでは説明を繰返さない。
【００７８】
直交検波器６より出力されるベースバンド信号ｘ（ｔ）、振幅計算部９２より出力される振幅Ａ_x（ｔ）および位相計算部９８より出力される位相φ_x（ｔ）との間には、次式（３１）に示される関係が成立する。振幅計算部９２は、次式（３２）により振幅Ａ_x（ｔ）を算出し、位相計算部９８は、次式（３３）により位相φ_x（ｔ）を算出する。
【００７９】
複素ベースバンド信号化部１００は、次式（３４）に従い、ベースバンド信号ｙ（ｔ）を計算する。
【００８０】
直交検波器２２より出力されるベースバンド信号ｚ（ｔ）、振幅計算部１１８より出力される振幅Ａ_z（ｔ）および位相計算部１２０より出力される位相φ_z（ｔ）との間には、次式（３５）に示される関係が成立する。振幅計算部１１８は、次式（３６）により振幅Ａ_z（ｔ）を算出し、位相計算部１２０は、次式（３７）により位相φ_z（ｔ）を算出する。
【００８１】
【数１１】

【００８２】
ＳＶＭ９４は、次式（３８）および（３９）に従って、ベースバンド信号ｙ（ｋＴ）の振幅を求める。
【００８３】
【数１２】

【００８４】
振幅に関しては、実施の形態１と同様の処理を行なう。
ＳＶＭ１１２は、次式（４０）〜（４３）に従って、Ａ′_ys（ｋＴ）を出力する。また、係数更新部１０４は、Ａ_ys（ｔ）とＡ′_ys（ｔ）との誤差が最小になるようにα_iを制御する。α_iは次式（４４）および（４５）で表わされる最適化問題を解くことにより得られる。
【００８５】
【数１３】

【００８６】
位相に関しては、以下に示すような処理が行なわれる。ＳＶＭ１１４は、次式（４６）〜（４９）に従って、Δφ_z（ｋＴ）を出力する。また、係数更新部１０８は、φ_ys（ｔ）とφ′_ys（ｔ）との誤差が最小になるように、β_iを制御する。
【００８７】
【数１４】

【００８８】
β_iは以下に示す更新手順（Ｓｔｅｐ０）〜（Ｓｔｅｐ５）に従って、係数更新部１０８で求められる。
【００８９】
（Ｓｔｅｐ０）
ｎ＝１、β_n＝１，φ′_y（ｐ_i）＝０，Δφ_z＝０とおく。
【００９０】
（Ｓｔｅｐ１）
学習データＶ＿Ａ_znおよびφ_s（ｐ_n）を入手する。ただし、Ｖ＿Ａ_znおよびφ_s（ｐ_n）は、次式（５０）および（５１）で表わされる。
【００９１】
【数１５】

【００９２】
（Ｓｔｅｐ２）
推定誤差ｅ（ｐ_n）が、減算器１１０において、次式（５２）および（５３）にしたがって計算される。
【００９３】
ｅ（ｐ_n）＝φ_ys（ｐ_n）−φ′_ys（ｐ_n） …（５２）
φ′_ys（ｐ_n）＝φ_z（ｐ_n）＋Δφ_z …（５３）
ｅ（ｐ_n）≦εであれが、β_n＝０，ｎ←ｎ＋１として（Ｓｔｅｐ１）に戻る。それ以外の場合には、（Ｓｔｅｐ３）に進む。
【００９４】
（Ｓｔｅｐ３）
係数β_nが次式（５４）に従って計算される。
【００９５】
【数１６】

【００９６】
（Ｓｔｅｐ４）
係数β_i，ｉ＝１，．．．，ｎの摂動分を次式（５５）に従い計算する。ただし、β_i＝０のデータは更新しない。
【００９７】
Δβ_i＝η_i｛ｅ_i−εｓｉｇｎ（β_i）｝ …（５５）
ｅ_iは前回計算した値を以下のルールで更新して用いる。
【００９８】
ｅ_iの初期値はｙ_s（ｐ_i）とする。
ｅ_i←ｅ_i−β_iＫ（Ｚ＿Ａ_zi，Ｚ＿Ａ_zi）， ｉ＝１，．．．，ｎ
…（５６）
η_i＝１／Ｋ（Ｒｅ｛Ｖ＿Ａ_zi，Ｖ＿Ａ_zi｝） （５７）
（Ｓｔｅｐ５）
次式（５８）を計算する。
【００９９】
【数１７】

【０１００】
式（５８）の計算の結果、Ｗ_n＞Ｗ_n-1であれば、係数β_iに摂動分Δβ_iを加算し、係数β_iを更新し、ｎを１つインクリメントし、（Ｓｔｅｐ１）に戻る。それ以外の場合には、係数β_iを更新することなく、係数β_iに０を代入し、Ｗ_nにＷ_n-1を代入し、ｎを１つインクリメントして（Ｓｔｅｐ１）に戻る。
【０１０１】
ＳＶＭ９６は、以上の処理により得られた係数β_iを用いて、次式（５９）の計算を行なう。
【０１０２】
【数１８】

【０１０３】
［実施の形態５］
図６を参照して、本発明の実施の形態５に係る前置型非線形歪補償器は、無線周波数信号Ｓ_INを中間周波数信号に変換するダウンコンバータ２と、ダウンコンバータ２に接続され、ダウンコンバータ２の出力する中間周波数信号をサンプリングしてデジタル信号（中間周波数信号Ｘ（ｔ））に変換するＡ／Ｄ変換器４と、Ａ／Ｄ変換器４に接続され、後述する増幅器１６の逆特性を有する関数を用いて、中間周波数信号Ｘ（ｔ）を中間周波数信号Ｙ（ｔ）に変換する７２と、ＳＶＭ７２に接続され、変換後の中間周波数信号Ｙ（ｔ）をアナログ信号に変換するＤ／Ａ変換器１２とを含む。
【０１０４】
前置型非線形歪補償器は、さらに、Ｄ／Ａ変換器１２に接続され、アナログの中間周波数信号を無線周波数信号Ｓ₀に変換するアップコンバータ１４と、アップコンバータ１４に接続され、無線周波数信号Ｓ₀を増幅して、出力信号Ｓ_OUTを得る増幅器１６と、増幅器１６に接続され、出力信号Ｓ_OUTを中間周波数信号に変換するダウンコンバータ１８と、ダウンコンバータ１８に接続され、中間周波数信号をデジタル信号（中間周波数信号Ｚ（ｔ））に変換するＡ／Ｄ変換器２０とを含む。
【０１０５】
前置型非線形歪補償器は、さらに、Ａ／Ｄ変換器２０および後述する係数更新部７８に接続され、Ａ／Ｄ変換器２０の出力する中間周波数信号Ｚ（ｔ）と、学習に用いた中間周波数信号ベクトルＺｉ（ｉ＝１，．．．，ｎ）とに基づいて、ＳＶＭ７２より出力される中間周波数信号Ｙ′_s（ｔ）を出力するＳＶＭ７４と、ＳＶＭ７２およびＳＶＭ７４に接続され、中間周波数信号Ｙ′_s（ｔ）と中間周波数信号Ｙ（ｔ）との差分を計算する減算器７６と、Ａ／Ｄ変換器４およびＡ／Ｄ変換器２０に接続され、中間周波数信号Ｘ（ｔ）およびＺ（ｔ）に基づいて、信号ベクトルおよびサポートベクトルのサイズｍと増幅器１６による遅延時間ｄとを求めるｍ・ｄ決定部８０と、減算器７６、ＳＶＭ７２、Ａ／Ｄ変換器２０およびＳＶＭ７４に接続され、減算器７６の出力、ＳＶＭ７２の出力、および中間周波数信号Ｚ（ｔ）に基づいてＳＶＭ７２における、増幅器１６の逆特性を有する関数の係数を更新する係数更新部７８とを含む。
【０１０６】
ここで、中間周波数信号Ｘ（ｔ）、Ｙ（ｔ）およびＺ（ｔ）は、中間周波数をｆ_IFとしたとき、次式（６０）〜（６２）で表わされる。ここで、ｘ（ｔ）、ｙ（ｔ）およびｚ（ｔ）は、実施の形態１で説明したベースバンド信号に相当する。
【０１０７】
Ｘ（ｔ）＝Ｒｅ｛ｘ（ｔ）ｅｘｐ（ｊ２πｆ_IFｔ）｝ …（６０）
Ｙ（ｔ）＝Ｒｅ｛ｙ（ｔ）ｅｘｐ（ｊ２πｆ_IFｔ）｝ …（６１）
Ｚ（ｔ）＝Ｒｅ｛ｚ（ｔ）ｅｘｐ（ｊ２πｆ_IFｔ）｝ …（６２）
実施の形態１に係る前置型非線形歪補償器は、中間周波数信号を直交変調し、ベースバンド信号に変換した後、ベースバンド信号に基づいて処理を行なっている。本実施の形態に係る前置型非線形歪補償器は、直交変調は行なわず、中間周波数信号のまま取扱いを行なうものである。しかし、その基本となる考え方は実施の形態１と同様である。このため、式の違いのみを説明し、その他の詳細な説明はここでは繰り返さない。
【０１０８】
ＳＶＭ７４は、以下の式（６３）〜（６６）に従って、Ｙ′_s（ｔ）を出力する。また、更新係数部７８は、Ｙ_s（ｔ）とＹ′_s（ｔ）との誤差が最小になるようにα_iを制御する。α_iは次式（６７）および（６８）で表わされる最適化問題を解くことにより得られる。
【０１０９】
【数１９】

【０１１０】
ＳＶＭ７２は、次式（６９）に従って、中間周波数信号Ｙ（ｋＴ）は計算される。
【０１１１】
【数２０】

【０１１２】
今回開示された実施の形態はすべての点で例示であって制限的なものではないと考えられるべきである。本発明の範囲は上記した説明ではなくて特許請求の範囲によって示され、特許請求の範囲と均等の意味および範囲内でのすべての変更が含まれることが意図される。
【０１１３】
【発明の効果】
ノイズフロアが低く、簡単な構成で高速処理可能な前置型非線形歪補償器を提供することができる。また、広帯域信号の補償も可能である。
【図面の簡単な説明】
【図１】 実施の形態１に係る前置型非線形歪補償器のハードウェア構成を示すブロック図である。
【図２】 係数更新部における処理のフローチャートである。
【図３】 実施の形態２に係る前置型非線形歪補償器のハードウェア構成を示すブロック図である。
【図４】 実施の形態３に係る前置型非線形歪補償器のハードウェア構成を示すブロック図である。
【図５】 実施の形態４に係る前置型非線形歪補償器のハードウェア構成を示すブロック図である。
【図６】 実施の形態５に係る前置型非線形歪補償器のハードウェア構成を示すブロック図である。
【符号の説明】
２，１８ ダウンコンバータ、４，２０ Ａ／Ｄ変換器、６，２２ 直交検波器、８，２８ ＳＶＭ、１０ 直交変調器、１２ Ｄ／Ａ変換器、１４ アップコンバータ、１６ 増幅器、２４，２８ 実数部抽出器、３０ 減算器、３２ 係数更新部。[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a front-end nonlinear distortion compensator, and more particularly to a front-end nonlinear distortion compensator using a support vector machine (hereinafter referred to as “SVM”) for regression analysis.
[0002]
[Prior art]
Conventionally, in order to realize nonlinear distortion compensation of a communication system, a pre-type nonlinear distortion compensator using various digital circuits has been created. There are two types of predistortion type non-linear distortion compensators using digital circuits, one using a look-up table and one estimating an inverse function of an amplifier.
[0003]
A pre-distortion distortion compensator using a look-up table divides the amplifier characteristics into small intervals with respect to the amplitude value of the input signal, and stores the representative values in the interval in a table to correspond to the transmission signal The output value is multiplied by the compensation value of the table to be output.
[0004]
On the other hand, a pre-linear distortion compensator that estimates the inverse function of an amplifier is called a pth-order-predistorter. This front-end nonlinear distortion compensator approximates the inverse function of a signal converter with a complex polynomial. The complex polynomial is derived from the most common Volterra series as a non-linear function expression format, and distortion compensation is performed.
[0005]
[Problems to be solved by the invention]
However, in the front-end nonlinear distortion compensator using a look-up table, the values in the table take discrete values, and the floor noise increases due to the discontinuity of the values. In order to eliminate the floor noise, it is necessary to increase the number of data. However, there is a problem that increasing the number of data increases the memory capacity.
[0006]
In the pth-order-predistorter, it is necessary to increase the order of the complex polynomial in order to increase the calculation accuracy. For this reason, there is a problem that it takes time to calculate and the solution is difficult to converge.
[0007]
Furthermore, for a wideband signal, both the look-up table and the pth-order-predistorter need to compensate for the nonlinearity of the amplitude and phase for each frequency in the signal band, so that hardware implementation is difficult. There's a problem.
[0008]
The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a front-end non-linear distortion compensator capable of wide-band distortion compensation with a simple configuration.
[0009]
Another object of the present invention is to provide a front-end nonlinear distortion compensator that can keep the noise floor low.
[0010]
Still another object of the present invention is to provide a pre-set nonlinear distortion compensator capable of high-speed processing.
[0011]
[Means for Solving the Problems]
A non-linear distortion compensator according to an aspect of the present invention includes a conversion support vector machine that converts an input signal using an inverse characteristic of a signal converter and supplies the converted signal to the signal converter; In the support vector machine for conversion based on the signal vector constituted by the conversion output signal sequence received from the conversion output signal converted by the converter and the signal vector constituted by the conversion output signal sequence extracted in the past. A support vector machine for estimation that estimates the input signal after conversion, and connected to the support vector machine for conversion and the support vector machine for estimation, and converted based on the output of the support vector machine for conversion and the output of the support vector machine for estimation Coefficient updater for updating coefficients of support vector machine for estimation and support vector machine for estimation Including the door.
[0012]
The inverse characteristics of the signal converter are estimated using an estimation support vector machine. Based on the output signal of the signal converter and the signal output from the conversion support vector machine, the coefficient updating means updates the coefficients of the conversion support vector machine and the estimation support vector machine. Since this configuration does not require a look-up table, the noise floor can be kept low, and a front-end nonlinear distortion compensator can be provided with a simple configuration. Further, the support vector machine is simpler to calculate and has a faster convergence time than the pth-order-predistorter. Therefore, it is possible to provide a front-end nonlinear distortion compensator capable of high-speed processing. Furthermore, distortion compensation is possible without requiring a special circuit (system) for a broadband signal.
[0013]
Preferably, the coefficient updating unit receives the conversion output signal, calculates a coefficient for a vector configured by the newly received conversion output signal sequence, and calculates the coefficient for the vector configured by the conversion output signal sequence input in the past. For the coefficient, only the perturbation is calculated, and the coefficient for the vector constituted by the conversion output signal sequence input in the past is updated.
[0014]
More preferably, the coefficient updating means is connected to the conversion support vector machine, and is connected to the first real part extraction means for extracting the real part of the output of the conversion support vector machine and the signal converter, and the signal A second real part extracting means for extracting the real part of the output of the converter, connected to the first and second real part extracting means, and based on the outputs of the first and second real part extracting means; And means for updating the coefficients of the transform support vector machine and the estimation support vector machine.
[0015]
More preferably, the pre-linear distortion compensator is further connected to the coefficient updating means, and based on the signal sequence of the input signal and the converted output signal sequence, the size of the signal vector and the support vector and the delay by the signal converter Means for calculating the time.
[0016]
More preferably, the support vector machine for conversion uses the inverse characteristic of the signal converter to convert the real part of the input signal and supplies the converted signal to the signal converter. An imaginary part conversion support vector machine that converts the imaginary part of the input signal using the inverse characteristics of the signal converter and supplies the converted signal to the signal converter. The support vector machine for estimation receives a real part of the converted output signal converted by the signal converter, and a signal vector constituted by a converted output signal sequence and a signal vector constituted by a converted output signal sequence extracted in the past. Based on the above, a support vector machine for real part estimation for estimating the input signal after conversion in the support vector machine for real part conversion, and an imaginary part of the conversion output signal converted by the signal converter, and receiving a conversion output signal sequence An imaginary part estimation support vector that estimates an input signal after conversion in an imaginary part conversion support vector machine based on a signal vector composed of the above and a signal vector composed of a conversion output signal sequence extracted in the past Including machines. The coefficient updating means is connected to the support vector machine for real part conversion and the support vector machine for real part estimation, and based on the output of the support vector machine for real part conversion and the output of the support vector machine for real part estimation, the real part conversion Real number coefficient updating means for updating the coefficients of the support vector machine and the real part estimation support vector machine, and the support vector machine for imaginary part conversion and the support vector machine for imaginary part estimation and support for imaginary part conversion Imaginary part coefficient updating means for updating the coefficients of the imaginary part conversion support vector machine and the imaginary part estimation support vector machine based on the vector machine output and the imaginary part estimation support vector machine output.
[0017]
More preferably, the conversion support vector machine is connected to the first amplitude calculator that calculates the amplitude of the input signal, the first phase calculator that calculates the phase of the input signal, and the first amplitude calculator. , Connected to the amplitude conversion support vector machine for converting the amplitude of the input signal into the amplitude of the signal after conversion, the first amplitude calculation unit and the first phase calculation unit, and the first amplitude calculation unit and the first phase Is connected to a support vector machine for phase conversion for obtaining the phase of the signal after conversion based on the output of the phase calculation unit, a support vector machine for amplitude conversion and a support vector machine for phase conversion, and a support vector machine for amplitude conversion and And a complex baseband signal converting unit for obtaining a converted signal based on the output of the support vector machine for phase conversion. The estimation support vector machine includes a second amplitude calculation unit for calculating the amplitude of the converted output signal converted by the signal converter, and a second phase calculation for calculating the phase of the converted output signal converted by the signal converter. And a signal vector that is connected to the second amplitude calculation unit and receives the output of the second amplitude calculation unit and is composed of the amplitude sequence of the converted output signal, and the amplitude sequence of the converted output signal extracted in the past And connected to an amplitude estimation support vector machine for estimating an amplitude of a signal after conversion in the amplitude conversion support vector machine, a second amplitude calculation unit, and a second phase calculation unit based on the configured signal vector. Receiving the outputs of the second amplitude calculation unit and the second phase calculation unit, the signal vector composed of the phase sequence of the converted output signal and the phase sequence of the converted output signal extracted in the past. That signal based on the vector, and a phase estimation support vector machines for estimating the phase of the signal after conversion in the phase conversion support vector machine. The coefficient updating means is connected to the support vector machine for amplitude conversion and the support vector machine for amplitude estimation, and based on the output of the support vector machine for amplitude conversion and the output of the support vector machine for amplitude estimation, Amplitude coefficient updating means for updating the coefficient of the support vector machine for amplitude estimation, and the support vector machine for phase conversion and the support vector for phase estimation connected to the support vector machine for phase conversion and the support vector machine for phase estimation Phase coefficient updating means for updating the coefficients of the phase conversion support vector machine and the phase estimation support vector machine based on the output of the machine.
[0018]
More preferably, the input signal, the output of the conversion support vector machine, and the output of the signal converter are intermediate frequency signals.
[0019]
More preferably, the signal converter is an amplifier.
[0020]
DETAILED DESCRIPTION OF THE INVENTION
[Embodiment 1]
Referring to FIG. 1, a pre-linear distortion compensator according to an embodiment of the present invention includes a radio frequency signal S. _IN Is converted to an intermediate frequency signal, an A / D converter 4 connected to the down converter 2 for sampling the intermediate frequency signal output from the down converter 2 and converting it into a digital signal, and an A / D converter 4 and a quadrature detector 6 for converting the sampled intermediate frequency signal to a baseband signal x (t), and a quadrature detector 6 connected to the quadrature detector 6, and using the inverse characteristic of the amplifier 16, the baseband signal x ( an SVM (support vector machine) 8 that converts t) into a baseband signal y (t), a quadrature modulator 10 that is connected to the SVM 8 and converts the converted baseband signal y (t) into an intermediate frequency signal; A D / A converter 12 connected to the quadrature modulator 10 and converting an intermediate frequency signal output from the quadrature modulator 10 into an analog signal.
[0021]
The front-end nonlinear distortion compensator is further connected to the D / A converter 12 and converts the analog intermediate frequency signal to the radio frequency signal S. ₀ And an output signal S output from the amplifier 16 and connected to the amplifier 16. _OUT Is converted to an intermediate frequency signal, a down converter 18 connected to the down converter 18, an A / D converter 20 converting the intermediate frequency signal into a digital signal, and an A / D converter 20 connected to A quadrature detector 22 for converting to a band signal z (t), and a real part extractor 24 connected to the quadrature detector 22 for extracting a real part Re {z (t)} of the baseband signal z (t). Including.
[0022]
The front-end nonlinear distortion compensator is further connected to a real part extractor 24 and a coefficient updating part 32 described later, and used for learning, a real part Re {z (t)} output from the real part extractor 24. Based on the baseband signal vector zi (i = 1,..., N), the SVM 26 that outputs the estimated value Re {y ′ (t)} of the real part of the baseband signal output from the SVM 8, and the SVM 8 To the real part extractor 28 for extracting the real part Re {y (t)} of the baseband signal y (t), and connected to the real part extractor 28 and the SVM 26 to estimate the real part of the baseband signal A subtractor 30 for calculating a difference between the value Re {y ′ (t)} and the real part Re {y (t)} of the baseband signal y (t), a subtractor 30, a real part extractor 28, and SVM8. , Connected to SVM 26 and real part extractor 24 A coefficient updating unit that updates the coefficients of the SVM 8 and the SVM 26 that realize the inverse characteristics of the amplifier 16 based on the output of the subtracter 30, the output of the SVM 26, the output of the real part extractor 28, and the output of the real part extractor 24. 32.
[0023]
The quadrature detector 6 is realized in analog, the output of the down converter 2 is received by the quadrature detector 6, the output of the quadrature detector 6 is converted to digital by the A / D converter 4, and the output is supplied to the SVM 8. You may be the structure to do. Further, the quadrature modulator 10 is realized in analog, the output of the SVM 8 is received by the D / A converter 12, the output of the D / A converter 12 is received by the quadrature modulator 10, and the output of the quadrature modulator 10 is up-converted. 12 may be provided. Further, the quadrature detector 22 is realized in analog, the output of the down converter 18 is received by the quadrature detector 22, the output of the quadrature detector 22 is received by the A / D converter 20, and the output of the A / D converter 20 is received. The configuration may be such that the real part extractor 24 is supplied.
[0024]
S _IN , S ₀ And S _OUT Are represented by formulas (1) to (3), respectively.

x (t), y (t) and z (t) are respectively the band signals S _IN , S ₀ And S _OUT The complex envelope (baseband signal). j represents an imaginary number. f _RF Represents a radio frequency. t represents time.
[0025]
[Operation principle of nonlinear distortion compensator]
P = | y (t) | ² Assuming that the input / output characteristics in the baseband region of the amplifier 16 is f (P) with f being a complex function, the relationship between the baseband signals y (t) and z (t) is expressed by the following equation (4). Can be expressed as follows.
[0026]
In general, a nonlinear distortion compensator aims to obtain an input / output characteristic g (P) that satisfies the relationship of the following expression (5) with respect to f (P). If the input / output characteristic g (P) can be obtained, the following equations (6) and (7) hold, and S _IN Against S _OUT Can be linearized.
[0027]
[Expression 1]

[0028]
In the present embodiment, g (P) is estimated by SVM 26 and coefficient updating unit 32. Specifically, the coefficient of the function of the SVM 26 is updated so that the error e (t) between y (t) and y ′ (t) estimated from z (t) is minimized. If this calculation converges, the SVM 26 has the inverse characteristics of the amplifier 16. For this reason, the nonlinearity of the amplifier 16 can be compensated by giving the same coefficient to the SVM 8 as the SVM 26.
[0029]
[Inverse function calculation]
In the present embodiment, the inverse characteristic g (P) of the input / output characteristic of the amplifier 16 is estimated using a support vector machine. The support vector machine itself is a well-known technique that has been extensively studied mainly in the case of application to classification problems such as the field of pattern recognition. However, there is no case where the support vector machine is applied to the regression problem as in the present invention.
[0030]
The conditions for calculation are summarized below.
Calculations for nonlinear distortion compensation are performed on discrete signals rather than time continuous signals. When hardware is used, an A / D converter is used to convert a continuous signal into a discrete signal, and after the calculation, the discrete signal is restored to a continuous signal by the A / D converter.
・ Signal processing is performed not on the band system but on the baseband system or the intermediate frequency signal. In the case of the baseband system, the amplifier 16 can be regarded as a system that receives the baseband signal y (t) equivalently and outputs z (t).
[0031]
[Description of operation of SVM 26]
Output y ′ of SVM 26 at time kT (T: sampling interval of A / D conversion and D / A conversion) _s (KT) is obtained by the following equation (8).
[0032]
[Expression 2]

[0033]
Here, m is the size of a signal vector and a support vector described later, n is the number of data vectors used for learning, and p _i Is the time when the i-th learning data vector is acquired. Learning data acquisition timing p _i Is determined experimentally from system performance. z _i (I = 1,..., N) is a data vector for learning the input / output characteristics of the amplifier 16. α _i Is z _i Is a weighting coefficient (real number) corresponding to. In general, α _i When is zero, the corresponding z _i Does not need to be added to the calculation and can be discarded. Conversely, α _i When is non-zero, the corresponding z _i Is called a support vector. Also, non-zero α _i Is called the dimension of the support vector machine. For example, non-zero α _i If the number of is 73, the number of dimensions of the support vector machine is 73. Furthermore, the number of dimensions does not increase beyond a certain value even if the learning data is increased. This upper bound value is called a VC (Vapnik and Chervonenkis) dimension.
[0034]
K (V_x, V_y) is a function called an inner product kernel. In the present embodiment, Equation (8), which is a linear sum of inner product kernels, is expressed as an appropriate α. _i Is selected to meet the inverse characteristic g (P) of the input / output characteristic of the amplifier 16. The variable prefix “V_” indicates that the variable is a vector. Such general problems that fit the input / output functions of the system to unknown functions are collectively called “regression problems”. The following equations (12) to (14) are examples of commonly used inner product kernels.
[0035]
[Equation 3]

[0036]
Which inner product kernel is most suitable for expressing the inverse function varies depending on the system. In practice, the best characteristic is selected after performing several tests before operation. Of course, the function other than the above can be used as long as the inverse function g (P) can be estimated better and the expected effect can be obtained.
[0037]
Further, in equation (8), only the real part of the complex number is extracted, and the output of the SVM 26 is calculated. This is to solve the problem that the support vector machine cannot handle complex numbers.
[0038]
[Description of operation of coefficient updating unit 32]
The coefficient updating unit 32 determines that the error e (t) between Re {y (t)} and Re {y ′ (t)} is minimized. _i To control. α _i Is obtained by solving the optimization problem expressed by the following equations (15) and (16).
[0039]
[Expression 4]

[0040]
However, a positive value C is a constant corresponding to a penalty for a sample that violates the constraint. ε is a margin for noise. Usually, both constants are obtained experimentally.
[0041]
The input / output characteristics of the amplifier 16 are considered to vary depending on the surrounding environment. For this reason, adaptively α _i Needs to be updated.
[0042]
The coefficient updating unit 32 performs α as follows: _i Update the value of.
(1) The coefficient updating unit 32 generates n-1 sets of data V_z. _i And coefficient α _i Suppose you have
(2) The coefficient updating unit 32 generates the nth data V_z. _n Get
(3) α ₁ ,. . . , Α _n-1 For, the coefficient is fixed and α _n Only the optimization problem expressed by the equation (15) is solved.
(4) α _n To correct the deviation (perturbation) from the optimal solution due to the addition of _i , (I = 1,..., N), error (perturbation) Δα _i Calculate
[0043]
As described above, by calculating only the perturbation of the coefficient due to the addition of new data, the amount of calculation can be reduced compared to the case where all the coefficients are recalculated.
[0044]
A specific example of the coefficient updating method is shown below. However, this specific example is an example of realization and does not necessarily have to be this way. In the following, T = 1 is set for the sake of simplicity. Even in this way, generality is not lost.
[0045]
Referring to FIG. 2, coefficient updating unit 32 calculates n = 1, α _n = 1, y ′ (p _n ) = 0 (S2). Learning data V_z _n And y _s (P _n ) Is obtained (S4). However, V_z _n And y _s (P _n ) Is expressed as follows: Here, d represents a delay time by the amplifier 16.
[0046]
[Equation 5]

[0047]
The subtractor 30 calculates the estimation error e (p _n ) Is calculated according to the following equation (19) (S6).
e (p _n ) = Re {y _s (P _n )}-Re {y ′ _s (P _n )} ... (19)
The coefficient updating unit 32 uses e (p _n ) And a margin ε for noise (see equation (15)) (S8). e (p _n ) ≦ ε (YES in S8), α _n 0 is substituted into n and n is incremented by 1 (S10). Thereafter, the process returns to S4.
[0048]
e (p _n )> Ε (NO in S8), the coefficient α _n Is calculated according to the following equation (20) (S12).
[0049]
[Formula 6]

[0050]
Next, the coefficient updating unit 32 calculates the coefficient α according to the following equation (21). _i , I = 1,. . . , N are calculated (S14). Where α _i For the coefficient where = 0, the perturbation is not calculated.
[0051]
Δα _i = Η _i {E _i -Εsign (α _i )} ... (21)
e _i Uses the previously calculated value updated by the following rules (1) to (2).
[0052]
(1) e _i The initial value of is y _s (P _i ).

And η _i Is obtained according to the following equation (23). The coefficient updating unit 32 calculates the following equation (24) (S16).
[0053]
[Expression 7]

[0054]
As a result of the calculation of Expression (24), W _n > W _n-1 If so (YES in S18), the coefficient α _i Perturbation component Δα _i And add coefficient α _i Is updated (S20). W _n ≦ W _n-1 If (NO in S18), α _n Substitute 0 for W _n To W _n-1 Is substituted (S22). After the process of S20 or S22, the coefficient updating unit 32 increments n by one and returns the process control to S4.
[0055]
[Description of SVM8 operation]
The SVM 8 uses the coefficient α updated by the coefficient updating unit 32. _i Is used to convert the baseband signal x (t) to the baseband signal y (t) according to the following equations (25) and (26). Im (V_x) represents the imaginary part of the complex number representation of the vector V_x.
[0056]
[Equation 8]

[0057]
In the above example, the support vector machine is used for correcting the output of the amplifier 16 having a nonlinear input / output characteristic whose input / output characteristic is unknown. However, the present invention is applicable only to the amplifier 16. is not. Generally, the present invention can be applied to a signal conversion circuit whose input / output characteristics are unknown.
[0058]
As described above, according to the present embodiment, a signal having the inverse characteristic of the amplifier 16 is estimated using the SVM 26. Based on the signal and the signal output from the SVM 8, the coefficient updating unit 32 updates the coefficient of the function having the inverse characteristic of the amplifier 16. Since this configuration does not require a look-up table, the noise floor can be kept low, and a front-end nonlinear distortion compensator can be provided with a simple configuration. Further, the support vector machine is simpler to calculate and has a faster convergence time than the pth-order-predistorter. Therefore, it is possible to provide a front-end nonlinear distortion compensator capable of high-speed processing.
[0059]
When a wideband signal is input, it is possible to compensate by appropriately setting the data vector size m even when the nonlinearity of the amplifier is narrower than the signal band.
[0060]
[Embodiment 2]
Referring to FIG. 3, the pre-linear distortion compensator according to Embodiment 2 of the present invention is the same as the pre-linear distortion compensator according to Embodiment 1 described with reference to FIG. The d determining unit 44 is added, and the coefficient updating unit 42 is used instead of the coefficient updating unit 32. Other components are the same as those in the first embodiment. For this reason, those descriptions are not repeated here.
[0061]
The m · d determination unit 44 is connected to the quadrature detection units 6 and 22. The coefficient updating unit is connected to the SVM 8, the real part extracting unit 28, the subtracter 30, the SVM 26, the real part extracting unit 24, and the m · d determining unit 44.
[0062]
The m · d determining unit 44 uses the evaluation function e represented by the following equation (27). _r The size m of the signal vector and the support vector and the delay time d by the amplifier 16 are obtained so that (k) is minimized. The coefficient updating unit 42 uses the m and d determined by the m, d determining unit 44 to use the coefficient α. _n Update. The operation of the coefficient updating unit 42 is the same as that shown in the first embodiment. Therefore, detailed description thereof will not be repeated here.
[0063]
e _r (K) = | z (t) −kx (t) | ² ... (27)
The SVM 8 obtains the baseband signal y (kT) according to the following equation (28).
[0064]
[Equation 9]

[0065]
[Embodiment 3]
Referring to FIG. 4, the front-end nonlinear distortion compensator according to Embodiment 3 of the present invention includes a radio frequency signal S. _IN Is converted to an intermediate frequency signal, an A / D converter 4 connected to the down converter 2 for sampling the intermediate frequency signal output from the down converter 2 and converting it into a digital signal, and an A / D converter 4 and a quadrature detector 6 for converting the sampled intermediate frequency signal to a baseband signal x (t), and a quadrature detector 6 connected to the quadrature detector 6 and the real part Re {x ( t)} to extract a real part extractor 52, an imaginary part extractor 56 connected to the quadrature detector 6 to extract an imaginary part Im {x (t)} of the baseband signal x (t), and an imaginary part An SVM 54 connected to the extractor 52 for converting Re {x (t)} into a real part Re {y (t)} of the baseband signal y (t) using an inverse characteristic of the amplifier 16 described later; In the imaginary part extractor 56 Connected to SVMs 58 and 58, which convert Im {x (t)} into an imaginary part Im {y (t)} of the baseband signal y (t) using the inverse characteristics of the amplifier 16. A quadrature modulator 10 that converts the converted baseband signal y (t) into an intermediate frequency signal, and a D / D that is connected to the quadrature modulator 10 and converts the intermediate frequency signal output from the quadrature modulator 10 into an analog signal. A converter 12.
[0066]
The front-end nonlinear distortion compensator is further connected to the D / A converter 12 and converts the analog intermediate frequency signal to the radio frequency signal S. ₀ And an output signal S output from the amplifier 16 and connected to the amplifier 16. _OUT Is converted to an intermediate frequency signal, a down converter 18 connected to the down converter 18, an A / D converter 20 converting the intermediate frequency signal into a digital signal, and an A / D converter 20 connected to A quadrature detector 22 that converts to a band signal z (t); a real part extractor 24 that is connected to the quadrature detector 22 and extracts a real part Re {z (t)} of the baseband signal z (t); And an imaginary part extractor 60 that is connected to the quadrature detector 22 and extracts the imaginary part Im {z (t)} of the baseband signal z (t).
[0067]
The front-end nonlinear distortion compensator is further connected to a real part extractor 24 and a coefficient updating unit 42 described later, and is used for learning with a real part Re {z (t)} output from the real part extractor 24. Based on the baseband signal vector zi (i = 1,..., N), the SVM 26 that outputs the estimated value Re {y ′ (t)} of the real part of the baseband signal output from the SVM 54, and the imaginary number Imaginary part Im {z (t)} output from the imaginary part extractor 60 and the baseband signal vector zi (i = 1,..., N). Are connected to the SVM 62 that outputs the estimated value Im {y '(t)} of the imaginary part of the baseband signal output from the SVM 58, and the estimated value Re of the real part of the baseband signal. {Y '( )} And a subtractor 30 for calculating the difference between the real part Re {y (t)} of the baseband signal y (t), and the estimated value Im {of the imaginary part of the baseband signal, connected to the SVM 58 and SVM 62. y ′ (t)} and a subtractor 64 for calculating the difference between the imaginary part Im {y (t)} of the baseband signal y (t) and the quadrature detectors 6 and 22 to be connected to the signal vector and support The m · d determining unit 44 for obtaining the vector size m and the delay time d by the amplifier 16 is connected to the subtractor 30, SVM 54, SVM 26, the real part extractor 24 and the m · d determining unit 44. Based on the output, the output of the SVM 26, the output of the SVM 54, the output of the real part extractor 24 and the output of the m · d determining unit 44, the coefficient of the function having the inverse characteristic of the amplifier 16 in the SVM 54 is updated. The number update unit 42 is connected to the subtractor 64, SVM 58, SVM 62, real part extractor 60, and m · d determination unit 44, and the output of the subtractor 64, the output of SVM 62, the output of SVM 58, and the real part extractor 60 Based on the output and the output of the m · d determining unit 44, a coefficient updating unit 66 that updates the coefficient of the function having the inverse characteristic of the amplifier 16 in the SVM 58 is included.
[0068]
The SVM 54 and the SVM 58 perform only the calculation for the real part and the imaginary part, respectively, among the calculations of the SVM 8. The calculation of SVM8 is the same as that described in the first embodiment. Therefore, detailed description thereof will not be repeated here.
[0069]
The SVM 62 obtains the estimated value Im {y ′ (t)} of the imaginary part of the baseband signal according to the same method as the estimated value Re {y ′ (t)} of the real part of the baseband signal in the SVM 26. The processing in the SVM 26 is as described in the first embodiment.
[0070]
The coefficient updating unit 66 determines that the coefficient updating unit 32 minimizes the error e (t) between Re {y (t)} and Re {y ′ (t)}. _i Is controlled so that the error between Im {y (t)} and Im {y ′ (t)} is minimized. _i To control. The processing in the coefficient updating unit 32 is as described in the first embodiment.
[0071]
As described above, according to the non-linear distortion compensator of this embodiment, the real part of the amplifier baseband output is corrected for the real part of the input signal, and the imaginary number of the amplifier baseband output is corrected for the imaginary part. Each part is used to estimate the inverse function with a separate support vector machine, and the real and imaginary parts of the input signal are corrected separately. With such a configuration, the compensation accuracy is improved.
[0072]
The SVM 54 obtains the real part of the baseband signal y (kT) according to the following equation (29), and the SVM 58 obtains the imaginary part of the baseband signal y (kT) according to the following equation (30).
[0073]
[Expression 10]

[0074]
[Embodiment 4]
Referring to FIG. 5, the front-end nonlinear distortion compensator according to Embodiment 4 of the present invention includes a radio frequency signal S. _IN Is converted to an intermediate frequency signal, an A / D converter 4 connected to the down converter 2 for sampling the intermediate frequency signal output from the down converter 2 and converting it into a digital signal, and an A / D converter 4 and a quadrature detector 6 for converting the sampled intermediate frequency signal to a baseband signal x (t), and an amplitude when the baseband signal x (t) is expressed in polar coordinates by being connected to the quadrature detector 6. A _x The phase φ when the baseband signal x (t) is expressed in polar coordinates by being connected to the amplitude calculation unit 92 that outputs (t) and the quadrature detector 6. _x The baseband signal x (t) is converted into the baseband signal y (t) using the inverse characteristic of the amplifier 16 described later, which is connected to the phase calculation unit 98 that outputs (t) and the amplitude calculation unit 92. The amplitude A when the baseband signal x (t) is expressed in polar coordinates. _x Amplitude A when the baseband signal y (t) is expressed in polar coordinates from (t) _y When the baseband signal x (t) is converted into the baseband signal y (t) using the inverse characteristic of the amplifier 16 connected to the SVM 94 for deriving (t) and the amplitude calculator 92, the baseband Phase φ when signal y (t) is expressed in polar coordinates _y Phase φ when the baseband signal x (t) is expressed in polar coordinates from (t) _x Value obtained by subtracting (t) Δφ _x Is connected to the SVM 96 and the phase calculation unit 98, and the output Δφ of the SVM 96 is derived. _x And output φ of the phase calculation unit 98 _x (T) as input, phase φ _y An adder 102 that outputs (t), and is connected to the SVM 94 and the adder 102, and the amplitude A _y (T) and phase φ _y A complex baseband signal generating unit 100 that generates a complex baseband signal y (t) from (t) and an orthogonal unit that is connected to the complex baseband signal converting unit 100 and converts the baseband signal y (t) into an intermediate frequency signal. A modulator 10 and a D / A converter 12 connected to the quadrature modulator 10 and converting an intermediate frequency signal output from the quadrature modulator 10 into an analog signal are included.
[0075]
The front-end nonlinear distortion compensator is further connected to the D / A converter 12 and converts the analog intermediate frequency signal to the radio frequency signal S. ₀ An up-converter 14 for conversion into a radio frequency signal S connected to the up-converter 14 ₀ Output signal S _OUT And an output signal S connected to the amplifier 16. _OUT Is converted to an intermediate frequency signal, a down converter 18 connected to the down converter 18, an A / D converter 20 converting the intermediate frequency signal into a digital signal, and an A / D converter 20 connected to A quadrature detector 22 for converting to a band signal z (t) and an amplitude A when connected to the quadrature detector 22 and representing the baseband signal z (t) in polar coordinates. _z An amplitude calculation unit 118 that calculates (t) and a phase φ when the baseband signal z (t) is expressed in polar coordinates by being connected to the quadrature detector 22. _z And a phase calculation unit 120 that calculates (t).
[0076]
The front-end nonlinear distortion compensator is further connected to the amplitude calculation unit 118, and the amplitude A _z From (t), the amplitude A when the baseband signal y (t) is expressed in polar coordinates. _y Estimated value A ′ of (t) _y The SVM 112 that calculates (t) and the amplitude calculation unit 118 are connected to the amplitude A _z From (t), the phase φ when the baseband signal y (t) is expressed in polar coordinates _y Estimated value φ ′ of (t) _y From (t) to phase φ _z A value Δφ obtained by subtracting (t) _z Is connected to the SVM 112 and the amplitude A _y From (t), amplitude A ′ _y The subtractor 106 for obtaining a value obtained by subtracting (t) is connected to the SVM 114 and the phase calculation unit 120, and the phase φ _z (T) and the value Δφ _z (T) and the phase φ ′ _y The adder 116 for obtaining (t) is connected to the adder 102 and the adder 116, and the phase φ _y From (t), phase φ ′ _y A subtractor 110 for obtaining a value obtained by subtracting (t); an m · d determining unit 44 connected to the quadrature detector 6 for obtaining the size m of the signal vector and the support vector and the delay time d by the amplifier 16; The d determination unit 44, the subtractor 106, and the amplitude calculation unit 118 are connected, and the output of the m · d determination unit 44, the output of the subtractor 106, and the amplitude A _z Based on (t), the coefficient updating unit 104 that updates the coefficient of the SVM 94, the m · d determining unit 44, the subtractor 110, and the phase calculating unit 120 are connected, and the output of the m · d determining unit 44, the subtracter 110 Output and phase φ _z And a coefficient updating unit 108 that updates the coefficient of the SVM 96 based on (t).
[0077]
The same components as those in the first embodiment are denoted by the same reference numerals. Since the operation is the same, the description will not be repeated here.
[0078]
Baseband signal x (t) output from the quadrature detector 6 and amplitude A output from the amplitude calculator 92 _x (T) and the phase φ output from the phase calculation unit 98 _x The relationship shown in the following equation (31) is established between (t) and (t). The amplitude calculation unit 92 calculates the amplitude A by the following equation (32). _x (T) is calculated, and the phase calculation unit 98 calculates the phase φ by the following equation (33). _x (T) is calculated.
[0079]
The complex baseband signal converting unit 100 calculates the baseband signal y (t) according to the following equation (34).
[0080]
Baseband signal z (t) output from the quadrature detector 22, amplitude A output from the amplitude calculator 118 _z (T) and the phase φ output from the phase calculation unit 120 _z The relationship shown in the following equation (35) is established between (t) and (t). The amplitude calculator 118 calculates the amplitude A by the following equation (36). _z (T) is calculated, and the phase calculation unit 120 calculates the phase φ by the following equation (37). _z (T) is calculated.
[0081]
## EQU11 ##

[0082]
The SVM 94 obtains the amplitude of the baseband signal y (kT) according to the following equations (38) and (39).
[0083]
[Expression 12]

[0084]
For the amplitude, the same processing as in the first embodiment is performed.
The SVM 112 performs A ′ according to the following equations (40) to (43). _ys (KT) is output. In addition, the coefficient update unit 104 _ys (T) and A ' _ys Α so that the error from (t) is minimized. _i To control. α _i Is obtained by solving the optimization problem expressed by the following equations (44) and (45).
[0085]
[Formula 13]

[0086]
For the phase, the following processing is performed. The SVM 114 uses Δφ according to the following equations (46) to (49). _z (KT) is output. In addition, the coefficient updating unit 108 _ys (T) and φ ' _ys Β so that the error from (t) is minimized. _i To control.
[0087]
[Expression 14]

[0088]
β _i Is obtained by the coefficient updating unit 108 according to the following update procedures (Step 0) to (Step 5).
[0089]
(Step 0)
n = 1, β _n = 1, φ ' _y (P _i ) = 0, Δφ _z = 0.
[0090]
(Step 1)
Learning data V_A _zn And φ _s (P _n ). However, V_A _zn And φ _s (P _n ) Is expressed by the following equations (50) and (51).
[0091]
[Expression 15]

[0092]
(Step 2)
Estimated error e (p _n ) Is calculated in the subtractor 110 according to the following equations (52) and (53).
[0093]
e (p _n ) = Φ _ys (P _n ) −φ ′ _ys (P _n ... (52)
φ ′ _ys (P _n ) = Φ _z (P _n ) + Δφ _z ... (53)
e (p _n ) ≦ ε, β _n = 0, n ← n + 1, and return to (Step 1). In other cases, the process proceeds to (Step 3).
[0094]
(Step 3)
Coefficient β _n Is calculated according to the following equation (54).
[0095]
[Expression 16]

[0096]
(Step 4)
Coefficient β _i , I = 1,. . . , N are calculated according to the following equation (55). However, β _i Data of = 0 is not updated.
[0097]
Δβ _i = Η _i {E _i -Εsign (β _i )}… (55)
e _i Uses the previously calculated value updated with the following rules.
[0098]
e _i The initial value of is y _s (P _i ).
e _i ← e _i -Β _i K (Z_A _zi , Z_A _zi ), I = 1,. . . , N
... (56)
η _i = 1 / K (Re {V_A _zi , V_A _zi }) (57)
(Step 5)
The following equation (58) is calculated.
[0099]
[Expression 17]

[0100]
As a result of the calculation of Expression (58), W _n > W _n-1 Then the coefficient β _i Perturbation component Δβ _i And add the coefficient β _i Is updated, n is incremented by 1, and the process returns to (Step 1). Otherwise, the coefficient β _i Without updating the coefficient β _i Substitute 0 for W _n To W _n-1 Is substituted, n is incremented by 1, and the process returns to (Step 1).
[0101]
The SVM 96 uses the coefficient β obtained by the above processing. _i The following equation (59) is calculated using
[0102]
[Formula 18]

[0103]
[Embodiment 5]
Referring to FIG. 6, the pre-linear nonlinear distortion compensator according to the fifth embodiment of the present invention includes a radio frequency signal S. _IN Is converted to an intermediate frequency signal, and A / D conversion is connected to the down converter 2 and samples the intermediate frequency signal output from the down converter 2 and converts it into a digital signal (intermediate frequency signal X (t)). The intermediate frequency signal X (t) is converted to an intermediate frequency signal Y (t) using a function connected to the converter 4 and the A / D converter 4 and having an inverse characteristic of the amplifier 16 described later, and an SVM 72 And a D / A converter 12 for converting the converted intermediate frequency signal Y (t) into an analog signal.
[0104]
The front-end nonlinear distortion compensator is further connected to the D / A converter 12 and converts the analog intermediate frequency signal to the radio frequency signal S. ₀ An up-converter 14 for conversion into a radio frequency signal S connected to the up-converter 14 ₀ Output signal S _OUT And an output signal S connected to the amplifier 16. _OUT Is converted to an intermediate frequency signal, and an A / D converter 20 is connected to the down converter 18 and converts the intermediate frequency signal into a digital signal (intermediate frequency signal Z (t)).
[0105]
The front-end non-linear distortion compensator is further connected to the A / D converter 20 and a coefficient updating unit 78 described later, and is used for the intermediate frequency signal Z (t) output from the A / D converter 20 and learning. The intermediate frequency signal Y ′ output from the SVM 72 based on the intermediate frequency signal vector Zi (i = 1,..., N). _s The SVM 74 that outputs (t) and the intermediate frequency signal Y ′ connected to the SVM 72 and the SVM 74 _s (T) is connected to a subtractor 76 for calculating the difference between the intermediate frequency signal Y (t) and the A / D converter 4 and the A / D converter 20, and the intermediate frequency signals X (t) and Z (t ) To determine the size m of the signal vector and the support vector and the delay time d by the amplifier 16, and is connected to the subtractor 76, the SVM 72, the A / D converter 20 and the SVM 74 for subtraction. And a coefficient updating unit 78 for updating the coefficient of the function having the inverse characteristic of the amplifier 16 in the SVM 72 based on the output of the generator 76, the output of the SVM 72, and the intermediate frequency signal Z (t).
[0106]
Here, the intermediate frequency signals X (t), Y (t) and Z (t) _IF Is expressed by the following equations (60) to (62). Here, x (t), y (t), and z (t) correspond to the baseband signals described in the first embodiment.
[0107]
X (t) = Re {x (t) exp (j2πf _IF t)} (60)
Y (t) = Re {y (t) exp (j2πf _IF t)} (61)
Z (t) = Re {z (t) exp (j2πf _IF t)} (62)
The front-end nonlinear distortion compensator according to Embodiment 1 performs an orthogonal frequency modulation on an intermediate frequency signal and converts it into a baseband signal, and then performs processing based on the baseband signal. The front-end nonlinear distortion compensator according to the present embodiment is not subjected to quadrature modulation, but is handled as an intermediate frequency signal. However, the basic idea is the same as in the first embodiment. For this reason, only the difference in the equations will be described, and other detailed description will not be repeated here.
[0108]
The SVM 74 determines Y ′ according to the following equations (63) to (66): _s (T) is output. In addition, the update coefficient unit 78 _s (T) and Y ' _s Α so that the error from (t) is minimized. _i To control. α _i Is obtained by solving the optimization problem expressed by the following equations (67) and (68).
[0109]
[Equation 19]

[0110]
The SVM 72 calculates the intermediate frequency signal Y (kT) according to the following equation (69).
[0111]
[Expression 20]

[0112]
The embodiment disclosed this time should be considered as illustrative in all points and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the description above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.
[0113]
【The invention's effect】
A pre-linear distortion compensator having a low noise floor and capable of high-speed processing with a simple configuration can be provided. In addition, broadband signal compensation is also possible.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a hardware configuration of a front-end nonlinear distortion compensator according to a first embodiment.
FIG. 2 is a flowchart of processing in a coefficient updating unit.
FIG. 3 is a block diagram showing a hardware configuration of a front-end nonlinear distortion compensator according to the second embodiment.
FIG. 4 is a block diagram showing a hardware configuration of a front-end nonlinear distortion compensator according to a third embodiment.
FIG. 5 is a block diagram illustrating a hardware configuration of a front-end nonlinear distortion compensator according to a fourth embodiment.
FIG. 6 is a block diagram showing a hardware configuration of a front-end nonlinear distortion compensator according to a fifth embodiment.
[Explanation of symbols]
2,18 down converter, 4,20 A / D converter, 6,22 quadrature detector, 8,28 SVM, 10 quadrature modulator, 12 D / A converter, 14 up converter, 16 amplifier, 24, 28 real number Part extractor, 30 subtractor, 32 coefficient update part.

Claims (8)

A conversion support vector machine that converts the input signal using the inverse characteristics of the signal converter and supplies the converted signal to the signal converter;
The conversion output signal converted by the signal converter is received, and based on the signal vector configured by the conversion output signal sequence and the signal vector configured by the conversion output signal sequence extracted in the past, the conversion output signal A support vector machine for estimation for estimating the input signal after conversion in the support vector machine;
The support vector machine for conversion and the support vector for estimation are connected to the support vector machine for conversion and the support vector machine for estimation, and based on the output of the support vector machine for conversion and the output of the support vector machine for estimation, And a coefficient non-linear distortion compensator including coefficient updating means for updating a coefficient of the machine. The coefficient updating means receives the converted output signal, calculates the coefficient for a vector constituted by the newly received converted output signal string, and is constituted by the converted output signal string input in the past 2. The pre-linear distortion compensation according to claim 1, wherein only the perturbation component is calculated for the coefficient for the vector, and the coefficient for the vector configured by the converted output signal sequence input in the past is updated. vessel. The coefficient updating means includes
A first real part extracting means connected to the conversion support vector machine for extracting a real part of the output of the conversion support vector machine;
A second real part extraction means connected to the signal converter for extracting the real part of the output of the signal converter;
Connected to the first and second real part extraction means, and updates the coefficients of the conversion support vector machine and the estimation support vector machine based on the outputs of the first and second real part extraction means And a non-linear distortion compensator according to claim 1. And means for calculating the size of the signal vector and the support vector and the delay time by the signal converter based on the signal sequence of the input signal and the converted output signal sequence, connected to the coefficient updating means. The front-end nonlinear distortion compensator according to claim 1, comprising: The support vector machine for conversion is
Using the inverse characteristics of the signal converter, the real part of the input signal is converted, and a support vector machine for real part conversion that supplies the converted signal to the signal converter;
An imaginary part conversion support vector machine that converts the imaginary part of the input signal using the inverse characteristics of the signal converter and supplies the converted signal to the signal converter;
The estimation support vector machine is
The real part of the converted output signal converted by the signal converter is received, based on the signal vector configured by the converted output signal sequence and the signal vector configured by the converted output signal sequence extracted in the past, A support vector machine for real part estimation for estimating an input signal after conversion in the support vector machine for real part conversion;
The imaginary part of the converted output signal converted by the signal converter is received, based on the signal vector configured by the converted output signal sequence, and the signal vector configured by the converted output signal sequence extracted in the past, An imaginary part estimation support vector machine for estimating an input signal after conversion in the imaginary part conversion support vector machine,
The coefficient updating means includes
The real part conversion is connected to the real part conversion support vector machine and the real part estimation support vector machine and based on the output of the real part conversion support vector machine and the output of the real part estimation support vector machine Real part coefficient update means for updating the coefficients of the support vector machine for use and the support vector machine for estimating the real part,
The imaginary part conversion is connected to the support vector machine for imaginary part conversion and the support vector machine for imaginary part estimation, and based on the output of the support vector machine for imaginary part conversion and the output of the support vector machine for imaginary part estimation, the imaginary part conversion 2. The pre-linear distortion compensator according to claim 1, further comprising: an imaginary part coefficient updating unit for updating a coefficient of the imaginary part estimation support vector machine and the imaginary part estimation support vector machine. The support vector machine for conversion is
A first amplitude calculator for calculating the amplitude of the input signal;
A first phase calculator for calculating the phase of the input signal;
An amplitude conversion support vector machine connected to the first amplitude calculator for converting the amplitude of the input signal into the amplitude of the converted signal;
A phase that is connected to the first amplitude calculator and the first phase calculator, and obtains the phase of the converted signal based on the outputs of the first amplitude calculator and the first phase calculator in the previous period A support vector machine for conversion,
A complex baseband that is connected to the support vector machine for amplitude conversion and the support vector machine for phase conversion, and obtains the converted signal based on outputs of the support vector machine for amplitude conversion and the support vector machine for phase conversion Including a signaling unit,
The estimation support vector machine is
A second amplitude calculator for calculating the amplitude of the converted output signal converted by the signal converter;
A second phase calculator for calculating the phase of the converted output signal converted by the signal converter;
A signal vector connected to the second amplitude calculation unit, receiving the output of the second amplitude calculation unit and configured by an amplitude sequence of the converted output signal, and an amplitude sequence of the converted output signal extracted in the past A support vector machine for amplitude estimation that estimates the amplitude of the signal after conversion in the support vector machine for amplitude conversion based on a configured signal vector;
Connected to the second amplitude calculation unit and the second phase calculation unit, receives outputs of the second amplitude calculation unit and the second phase calculation unit, and is constituted by a phase sequence of the converted output signal A phase estimation support vector machine for estimating a phase of a signal after conversion in the phase conversion support vector machine based on a signal vector and a signal vector constituted by a phase sequence of a conversion output signal extracted in the past; Including
The coefficient updating means includes
Connected to the support vector machine for amplitude conversion and the support vector machine for amplitude estimation, and based on the output of the support vector machine for amplitude conversion and the output of the support vector machine for amplitude estimation, the support vector machine for amplitude conversion and Amplitude coefficient updating means for updating coefficients of the support vector machine for amplitude estimation;
The phase conversion support vector machine and the phase estimation support vector machine connected to the phase conversion support vector machine and based on the output of the phase conversion support vector machine and the output of the phase estimation support vector machine, and 2. The pre-linear distortion compensator according to claim 1, further comprising phase coefficient updating means for updating a coefficient of the support vector machine for phase estimation. 2. The pre-linear distortion compensator according to claim 1, wherein the input signal, the output of the conversion support vector machine, and the output of the signal converter are intermediate frequency signals. The pre-linear distortion compensator according to claim 1, wherein the signal converter is an amplifier.

Images