CA2335433A1 - Method for transmitting digital data via a transmission channel subject to perturbations occurring in bursts - Google Patents

Abstract

The invention relates to a method for transmitting digital data via a transmission channel (2) subject to perturbations occurring in bursts.
According to said method a soft-decision decoding (9) taking into account an estimated signal quality is carried out in a receiver (3) after a demodulation (7). To determine signal quality a defined number of received symbols can be tested to determine whether they are situated within a defined distance from a constellation point. The number of symbols situated within said defined distance can be used as quality value for the next decoding. In this way a characteristic curve (16.1 to 16.3) with a bulge (17.1 to 17.3) is obtained.
Symbols with a low noise level are relatively overevaluated while those with a higher noise level are relatively underevaluated.

Description

Method for transmitting digital data via a transmission channel subject to perturbations occurring in bursts Technical field The invention relates to a method for transmitting digital data via a transmission channel which is subject to interference occurring in bursts, where a soft-decision decoding step taking into account an estimated signal quality is carried out in a receiver after a demodulation step. The invention also relates to a receiver circuit for carrying out the method.
Prior art The liberalization of telecommunications has led to a search for opportunities to provide broadband transmission via existing cables and media not primarily designed for data transmission. A particular area of interest is the use of the power cable grid, because all households are already connected to this grid, so to speak. In other words: the transmission medium exists and just needs to be used in a suitable manner.
The problem of electrical cables is that powerful pulsed interference (typically caused by switching processes) is present. Other interference sources (echoes, radiated interference etc.) also make it harder to transmit broadband signals. Greatly varying interference may also arise in radio channels.
Description of the invention The object of the invention is to specify a method of the type mentioned initially which provides improved detection in the presence of pulsed or bursty interference.
The object is achieved in a manner defined by the features of Claim 1. According to the invention, the reception quality is estimated using a circuit whose characteristic has a bulge (or S curve characteristic) which produces an overassessment with respect to a linear characteristic when there is a relatively low [sic] signal-to-noise ratio and produces a relative underassessment of the signal quality when there is a relatively high [sic] signal-to-noise ratio.
In the case of pulsed interference, this allows for the fact that the noise level may rise sharply for a brief period and that the reliability of the symbol estimates is then relatively low. In principle, the short, highly erroneous signal portions are selectively marked as "poor" for decoding.
The signal quality or the signal-to-noise ratio is estimated on the basis of the variance of the interference process within a prescribed window. In accordance with one preferred embodiment, this window has a length in the region of the shortest interference pulses which can be expected. The effect achieved by this is that estimation of the signal quality reacts sufficiently quickly to interference pulse bursts which occur.
Measurements show that, in the case of power supply cables, the interference pulses which occur have a length in the region of between several hundred nanoseconds and a few tens of microseconds. In this interference environment, window lengths of, typically, less than 3 ~.s (e.g. of 1 ~.s) have been found to be advantageous. For a symbol rate of, by way of example, 15 Ms/s, a window length of at least approximately N = 15 is optimum. A length which is all too short impairs the reliability of estimating the signal quality. On the other hand, with a length of, by way of example, N = 50, the reaction to short interference pulses is unsatisfactory. The window length should therefore be in the range from N = 3 to N = 25, particularly in the range from N = 5 to N = 15.
In accordance with one particularly preferred embodiment, to determine the signal quality, a prescribed (on the basis of the window length) number of received symbols are tested to determine whether they are situated within a predefined distance from one of the constellation points present. The number of symbols situated within the aforementioned distance is used as quality value for the subsequent soft-decision decoding.
The choice of distance used as threshold value depends on the modulation method, on the statistics for the interference pulses and, individually, on further parameters. In a first approximation, the distance can be chosen such that, with a signal-to-noise ratio of at least approximately 5 dB, the majority of the symbols in the particular window are situated within the prescribed distance from a constellation point from a statistical standpoint. This is because experiments have shown that the bulge in the characteristic should be situated below approximately 5 dB. Consequently, symbol estimations carried out at approximately 0 dB or less are qualified as "poor". On the other hand, with a signal-to-noise ratio of more than 10 dB, symbol estimation can be assumed to be reliable.
For a BPSK modulation method (BPSK = Binary Phase Shift Keying), a distance of D = 0.5 is particularly advantageous. Considered overall, this value can be used to achieve the best system performance. Of course, other values are not fundamentally unsuitable.
The implementation of the method according to the invention affects the receiver only. The transmitter circuits may be designed in a manner which is known per se.

In the receiver, a module which estimates the symbol quality on the basis of the method described is inserted between the demodulation stage and soft-decision decoding. The estimated quality value is used, together with the estimated symbol value, in the subsequent decoding to ascertain the transmitted symbol. Whether the module according to the invention is implemented in software or in hardware is of no consequence for the invention.
The detailed description below and the patent claims in their entirety reveal further advantageous embodiments and combinations of features of the invention.
Brief description of the drav~ings In the drawings used to explain the illustrative embodiment:
Figure 1 shows a block diagram to explain the invention;
Figure 2 shows a schematic illustration of a preferred embodiment of the invention;
Figure 3 shows examples of characteristic curves for various distances D;
Figures 4a to d show illustrations of the performance for different types of noise pulses.
In principle, identical parts are provided with the same reference symbols in the figures.
Methods of implementing the invention Figure 1 shows a transmission diagram with FEC
coding (FEC - Forward Error Correction) between a transmitter 1 and a receiver 3. The digital data from a data source 4 is coded in a coder 5 using an inherently known method to produce symbols, in order to permit receiver-end correction of transmission errors. The modulator 6 performs modulation appropriate to the transmission channel 2.
By way of example, the transmission channel 2 is formed by a radio transmission link or by cables in an electrical power distribution grid. It is assumed that said transmission channel is subject to interference (interference pulses, beats, echoes etc.) which varies over time. For the examples further below, it is assumed that the shortest interference pulses which can be expected have a length in the region of 1 ~s (or slightly less). In the absence of interference pulses, a signal-to-noise ratio of at least approximately 10 dB is assumed.
In the receiver 3, the received signal is demodulated in the demodulator 7 (which may also comprise an equalizer) and is decoded in the decoder 9 in order to ascertain the data which is to be output to the data sink 10. The decoder 9 is a soft-decision decoder operating on the basis of known principles. An essential part of the invention is the signal quality estimator 8, which ascertains the reliability of the symbols estimated by the demodulator 7 and provides a corresponding reliability value for the soft-decision decoding.
Figure 2 shows a block diagram of the signal quality estimator 8. In this case, a prerequisite is that symbols having a permanently prescribed constellation are produced in the transmitter 1. These symbols are generally transmitted in succession, but may also be transmitted simultaneously (e.g. at different frequencies). In the context of the statements below, a temporal sequence is assumed.
The noise process (which may also be based on beats) may be regarded as being random in the receiver.
The interference is adopted in bursty or pulsed form.
This means that it provides a "customary" noise power and that this noise power rises to a very high value for a short period from time to time, and then falls back to the "customary" value. Both the rise and the duration of the increase in noise power are random.
In accordance with the invention, the noise power is ascertained for each transmitted symbol.
Without any restriction of general nature, it can be . assumed that the signal power is known (e. g. on account of a measurement at the start of reception of a data block). What is sought, however, is not the instantaneous power of the interference, but rather the variance of the noise process at a given time.
Estimation of the noise power is based on a block or a window of N successive symbols. The length N
of the window is chosen on the basis of the expected noise process. With constant signal quality, greater window lengths result in greater accuracy, but in a slower reaction to changes in the signal quality. For PLC applications (PLC = Power Line Communication) with a symbol rate (symbols per second) of, for example, 15 Ms/s, a length of N = 15 has been found to be advantageous. This is because the shortest interference pulses to be expected are in the region of 1 ~s. If, by way of example, a symbol rate of 5 Ms/s were chosen, a length of N = 5 to N = 7 would be advantageous.
The noise level ascertained with the window is used for the symbol situated in the center of the window. If N is uneven, the result of the signal quality estimation depends on (N-1)/2 preceding and on (N-1)/2 succeeding demodulated symbols and on the symbol whose quality is to be estimated. (The length N
may also be even, in which case the estimation method is then not symmetrical, however. This has no negative effects for the performance of the method according to the invention, though.) The invention is primarily intended for situations in which the signal power is essentially constant and the noise power is variable. If the signal power likewise varies over time, but varies at a slower rate than the noise power, the invention may likewise be used, but an additional level of computation complexity is necessary to track the signal power.
The method illustrated in Figure 2 for determining a quality value may be illustrated as a nonlinear FIR filter (FIR = Finite Impulse Response):
The estimated symbols si are input into a shift register having N delay elements 11.1 to 11.N. The output of each delay element 11.1 to 11.N is forwarded to one computer 12.1 to 12.N in each case. The computers 12.1 to 12.N determine the distance between the respective symbol and the next constellation point.
In the case of BPSK modulation with the two constellation points +1 and -1, the first thing established, for example, is whether the magnitude value is greater than or less than 0. In the former case, the Euclidean distance Di from the constellation point +1 is calculated and in the latter case the corresponding distance from the constellation point -1 is calculated.
In the next step, the comparisons [sic] 13.1 to 13.N establish for each distance Di whether the respective distance Di is shorter than a firmly prescribed distance D. If the result of the comparison is positive, i.e. if the distance is shorter, then a value xi = 1 is output at the output of the respective comparator 13.1 to 13. N.
Finally, the adders 14.1 to 14.N-1 calculate the sum of all the xi values. On the basis of the above statements, it is clear that the resultant reliability 3 0 value x has the value range [ 0 , 1, ..., N] .
The parameter D is stipulated taking into account the modulation and coding method and the error rate which can be expected. D should be defined so as to be proportional to the amplitude of the signal received. For real-value bipolar modulation with a standard amplitude (BPSK) and convolutional coding at a rate of 1/2, a value D = 0.5 is preferred.

Calculation of the quality value entails a delay of (N-1)/2. For this reason, a delay element 15 of appropriate size needs to be inserted in the direct path between the demodulator 7 and the decoder 8.
Particular measures are required at the start and at the end of the symbol stream. By way of example, the first symbol has no predecessor. If the missing xi are initiated with zeros, then the quality of the first pair of symbols is underassessed, whereas, if they are initialized with the value of the first symbol, then the latter is overassessed.
In accordance with a preferred embodiment, the window is therefore shortened at the start and at the end of the symbol stream by, for example in the case of the first symbol, taking into account only this symbol and the (N+1 ) /2 succeeding symbols . In the case of the second and penultimate symbols, the quality is based on (N+3)/2 symbol values, for example (namely on the first and last, on the second and penultimate and on the succeeding (N+1)/2 symbols). In this regard, it should be noted that the result of the quality estimator should be scaled using a factor A/N (A denotes the number of valid symbols in the shift register (delay elements 11.1 to 11.N) [lacuna] .
Another option is to ignore all starting values for the quality estimator 8 which are not based on N
valid symbols. For the first and last (N+1)/2 symbols, there are no estimated values for the variance in the interference process in this case. These gaps could be filled by repeating the first and last valid estimated values, for example.
The general block diagram shown in Figure 2 may be simplified according to circumstances (e.g. if the demodulated symbols relate to real values and are bipolar). These simplifications will not be discussed in more detail at this point, however (since they have no significant influence on the result).

Figure 3 shows, by way of example, three different characteristic curves according to the invention (16.1 to 16.3). It should be pointed out that all three characteristic curves are monotone. Plotted on the abscissa is the output value from the signal quality estimator 8, and plotted on the ordinate is the signal-to-noise ratio (SNR). A window length of N = 15 and a BPSK method are assumed. The characteristic curves 16.1 to 16.3 correspond to different predefined distances: D = 0.25, D = 0.50, D = 0.75. Each characteristic curve has a bulge according to the invention 17.1 to 17.3 which is in the range between -5 to +5 dB SNR in each case. (The result of the bulge or S characteristic is that the sensitivity of the output signal in a certain range reacts less sensitively to changes in the SNR, which contributes to a lack of criticality of inaccuracies when measuring the variance in the interference process, i.e. relatively short window lengths.) The result of the S curve characteristic is an overassessment with respect to a linear comparative curve (cf. dashed curve 18 relating to characteristic curve 16.1, for example) when there is a relatively low [sic] (in terms of the operating point) signal-to-noise ratio, and a relative underassessment of the signal quality when there is a relatively high [sic]
signal-to-noise ratio.
It should be pointed out that the method according to the invention is particularly suitable for signal/noise ratios in the range from approximately 5 dB to 15 dB. As a result of a suitable choice of distance parameter D, the characteristic curves will be positioned such that the bulge is situated in the region of the critical SNR value (modulation and coding methods usually have a small critical region for the SNR: above this region, performance is good, whereas below it, it is poor). If the noise level rises sharply for a brief period, the "operating point" is temporarily shifted to the lower side or edge of the bulge. Under the boundary conditions on which Figure 3 is based, the characteristic curve 16.2 is the best suited.
Figures 4a to 4d show simulation results for packet-oriented data transmission using BPSK via the electrical supply cables. The data rate has been stipulated at 15 Ms/s. To combat channel interference, an equalizer and a convolutional code have been used.
The graphical illustrations show the packet error rate in the presence of pulsed (bursty) noise. A comparison is made between the performances with and without the quality estimation according to the invention. The length of the filter was N = 15 and the prescribed distance was stipulated as D = 0.5 In all simulations, the background noise was fixed ("customary" noise), and noise pulses of a prescribed length were injected. In each case, the abscissa shows the ratio between the level of the interference pulses and the level of the background noise in dB. Plotted on the ordinate is the rate of lost packets. The curves marked by "+" show the performance without the quality estimation according to the invention, and the curves marked by "x" show the performance using the invention.
In Figure 4a, the SNR is 25 dB and the interference pulses have a length of 1 acs. The advantage of the invention is striking: interference pulses having levels of above 35 dB with respect to the background noise can be tolerated. If the quality estimation according to the invention is dispensed with, a difference of only 20 dB can be accepted.
Figure 4b is likewise based on an SNR of 25 dB.
However, the interference pulses have a length of 10 us. As expected, the performance is worse with and without quality estimation (in general, there is more interference than in Figure 4a). Nevertheless, the invention achieves a benefit of 10 dB.

Figure 4c shows a situation with 25 dB SNR and with- interference pulses having a length of 50 ~.s.
Again, the overall noise power is greater and the performance in both cases is worse. The interference pulses are of such a length that the error correction coding cannot compensate for the data loss produced by the interference pulses. In this case, no benefit can be attained by the invention.
Finally, an SNR of 15 dB is taken as basis in Figure 4d. The interference pulses had a length of 1 acs. Again, a better performance is achieved using the quality estimation according to the invention.
(Although no benefit is achieved in the case of very small interference pulses of 10 dB and below, the overall performance of data transmission is so good under these circumstances that the perceivable difference is of no significance.) In summary, it should be stated that the invention permits the performance of known signal transmission methods to be improved when there is pulsed interference present on the transmission channel. In principle, the invention is suitable for any linear modulation method using FEC in which the coded symbols coincide with a discrete, preferably small number of constellation points.

Claims (10)

claims

1. Method for transmitting digital data via a transmission channel (2) which is subject to interference occurring in bursts, where a soft-decision decoding step (9) taking into account an estimated signal quality is carried out in a receiver (3) after a demodulation step (7), characterized in that the signal quality of the symbols present after the demodulation (7) is carried out [sic] using a quality assessment circuit whose characteristic curve (16.1) has a bulge (17.1) or S characteristic which produces an overassessment with respect to a linear characteristic curve when there is a relatively low noise level and produces an underassessment when there is a relatively high noise level.

2. Method according to Claim 1, characterized in that the signal quality is estimated on the basis of the variance of the interference process within a prescribed window of symbols, the window having a length (N) in the region of the shortest interference which can be expected on a statistical basis.

3. Method according to Claim 2, characterized in that the length (N) of the window is chosen to be in the region of less than 3 µs for a transmission channel formed by power supply cables.

4. Method particularly according to one of Claims 1 to 3, characterized in that, to determine the signal quality, a prescribed number of received symbols are tested to determine whether they are situated within a predefined distance (D) from a constellation point; and in that the number of symbols situated within the predefined distance is used as quality value for the subsequent decoding.

5. Method according to Claim 4, characterized in that the predefined distance is chosen such that, with a signal-to-noise ratio of at least 5 dB, the majority of the symbols are situated within the aforementioned distance from a constellation point from a statistical standpoint.

6. Method according to one of Claims 2 to 5, characterized in that the window has a length of 3 to 25 symbols, particularly of approximately 5 to 15 symbols.

7. Method according to one of Claims 1 to 6, characterized in that BPSK is chosen as modulation method and a predefined distance of 0.5 with respect to a standard amplitude is chosen.

8. Receiver circuit for carrying out the method according to Claim 1, having a demodulator (7) and a soft-decision decoder (9), characterized by a quality assessment circuit (8) whose characteristic curve (16.1) has a bulge (17.1) which produces an overassessment with respect to a linear characteristic curve when there is a relatively low noise level and produces an underassessment when there is a relatively high noise level.

9. Receiver circuit according to Claim 8, characterized in that the quality assessment circuit (8) comprises means for determining the distance (Di) between a prescribed number of received symbols and the next particular constellation point, means for testing whether the distance is shorter than a predefined distance (D), and means for determining the number of symbols situated within the aforementioned distance (D).

10. Receiver circuit according to one of Claims 8 and 9, characterized in that the demodulator (7) is in the form of a BPSK demodulator.