Method for identifying received symbols corrupted by burst noise and related device

This invention relates to digital communications and more particularly to a method of identification of data corrupted by burst noise and related device and software for implementing the method.

In digital communication systems, a problem yet to be addressed is the presence of relatively long sequences of erroneously recognized bits (bursts). These bursts, due to disturbances of transient nature that affect the communication channel degrading the signal-to-noise ratio, worsen the overall performance of digital communication systems. It may be important to locate bursts in received data streams to read them again or to decode them in a specifically designed way.

A known method of detecting whether a burst has been received includes monitoring the envelope function of the signal to be demodulated. This implies that bursts that do not significantly modify the envelope function may hardly be detected.

Hereinafter, reference is made to communication systems in which transmitted symbols are defined as a sequence of a certain number P of bits that identifies a value of a Galois Field 2 P , and in which at the demodulator side, the probability mass function (PMF) of symbols, i.e. the probability P(u=φ|Y 1 N ) that a generic symbol u is equal to a certain value φ of the Galois field (GF) conditioned by a certain vector of samples Y N has been received, is made available at the output of the demodulator. This calculation can be done regardless the kind of transmission used (QAM, PSK, On/Off, etc), or the channel characteristics (memory-less, ISI channel, etc.) using well known algorithms (for example, symbol based BCJR, as described in Wu Chang and J. R. Cruz, Fellow, “Optimal Channel Detection for Nonbinary Coded Partial Response Channels”, IEEE Transactions on Communications , vol. 57, no. 7, July 2009, symbol based soft output Viterbi algorithms (SOVA), etc.).

Burst noise worsen statistic figures of a received sequence of bits, thus increasing the probability of an erroneous recognition. The prior art burst detection algorithms are generally based on bit based analysis, rather than on symbol analysis, and are typically inadequate when demodulation is tailored on GF symbols. Among the relevant contributions in this field, the authors Mustafa N. Kaynak, Tolga M. Duma, and Erozan M. Kurtas of “Burst Error Identification for Turbo- and LDPC-Coded Magnetic Recording Systems,” IEEE Transactions on Magnetics , Vol. 40, no. 4, July 2004, 3087, consider several approaches to identify the burst. Some of them are based on evaluating distortion of the expected signal shape. This type of approach is defeated by bursts of an unexpected type that may disturb the signal in unpredictable and an unidentifiable way. Other approaches are based on constraints imposed by the error correction code. These approaches are generally more onerous because they generally require the introduction of a sort of syndrome checker to identify a potential burst. Finally, the authors of “Burst Error Identification for Turbo- and LDPC-Coded Magnetic Recording Systems,” identified above, suggest considering a reliability parameter of the decision based on the probability of bits as estimated by the demodulator (or detector). This is a more general approach.

In “Improving burst error tolerance of LDPC-centric coding systems in read channel,” IEEE Transactions on Magnetics , Vol. 46, no. 3, March 2010, 933-941, by N. Xie, T. Zhang and E. F. Haratsch, a so-called hybrid concatenated coding is proposed to determine the presence of at least a burst error in the symbols of a read word of symbols. This special and sophisticated coding scheme requires interaction of different codes and is relatively complex. Moreover, it is still based on bit-based operations, and therefore is inadequate for transmission based on GF symbols.

An effective and reliable real time identification method of received symbols corrupted by burst noise and a related device for implementing it are described. The method includes in comparing the logarithm of at least one of the entries of the probability mass function, sorted in ascending (or descending) order from the least (most) probable value up (down) to the most (least) probable value, of each received symbol with a respective threshold. The method also includes counting the number of logarithms that exceed the corresponding threshold and generating a corresponding score for each received symbol. The generated sequence of scores is processed with a moving average, and the result is compared with an output threshold. When the output threshold is surpassed, the currently received symbol may be identified as corrupted by burst noise.

According to an embodiment, the score may be equal to the counting of the number of logarithms that exceed the respective threshold. According to yet another embodiment, which may be particularly useful in transmission conditions under which the channel characteristics may fluctuate, the thresholds with which the logarithm normalized probability mass functions are compared, may be continuously updated at every symbol received. The method may be implemented with a software computer program or with a related hardware device.

FIG. 1 is a sample graph of the logarithm of the probability of erroneous assessment of the value of a Galois Field 2 3 assumed by a received symbol.

FIG. 2 is a schematic block diagram of a device architecture for an algorithm for assessing the presence of a burst corrupting a received symbol according to the present invention.

FIG. 3 is a functional block diagram illustrating the algorithm according to the present invention.

FIG. 4 is a log-linear graph of SFR vs. SNR obtained by the algorithm according to the present invention.

The method will be disclosed referring to a situation in which the symbols may assume one out of eight values of a Galois Field 2 3 , though the same observations hold true for any generic Galois Field 2 P . Any type of transmission modulation (QAM, PSK, On/Off, etc.) may be used, provided that the probability mass function (PMF) of the values of each symbol is available at the receiver side.

For each received symbol, the receiver outputs a list of probability values (PMF) that the received symbol corresponds to the possible values of the Galois Field. In FIG. 1 , the PMF of a received symbol, sorted from the most probable value (that in the example is 1) down to the least probable value (that in the example is 8), are shown. The line labeled “No burst” is the average unconditional sorted PMF (AUSPMF), in the logarithm domain, of a symbol received outside of burst noise, and the line labeled “Over burst” is the AUSPMF of a symbol received during burst noise. As expected, the in-burst received symbols show greater uncertainness than that of out-of-burst received symbols.

Therefore, if there is a (long) burst in a received stream of data, a behavior similar to the one described with the line labeled “Over burst” is expected for a subsequent (large) number of symbols. Therefore, according to the method, the PMF of values calculated at the receiver side for a received symbol are processed and compared with thresholds fixed according to the method for determining whether the received symbol is corrupted by burst noise.

In FIG. 2 , a block diagram of an apparatus according to an embodiment is illustrated. The logarithms of the normalized probability values (more briefly the LogPMF) log(P(u=φ|Y 1 N )/P MAX ) available at the output of the receiver for each received symbol, are sorted from the largest to the smallest value and are compared with respective thresholds PMF_THRE, determined, as will be explained hereinafter. A score corresponding to the number of LogPMF exceeding the respective thresholds is assigned to the received symbol. By repeating the above operations for each received symbol, a sequence S(i) of scores is generated, and by processing this sequence, the symbols corrupted by burst noise are identified.

The sequence is preferably processed with a moving average filter (MA FILTER), that generates a moving average of the input sequence S(i) which, on its turn, is compared with a respective output threshold METRIC THRESHOLD. When this last threshold is exceeded, the presence of burst noise may be flagged, and the received symbol may be recognized as corrupted by burst. Tests carried out showed that relatively outstanding results may be obtained by calculating the moving average on 16 consecutive values of the sequence S(i), though a different number of consecutive values may be used.

The value of the METRIC THRESHOLD may be fixed by a user with a trial and error procedure in order to reduce the number of false and missing recognition of symbols corrupted by burst noise. In practice, this threshold may be linked to the number of PMF terms (from 1 to 2 P −1) that make up the sequence of scores S(i).

The thresholds PMF_THRE with which the sorted PMF of each symbol are compared generally depend on the working conditions on which the symbols are received. Therefore, even if it may be theoretically possible to let the user fix them heuristically, this may impose an onerous task to accomplish.

According to an aspect of the method, an algorithm for automatically setting and continuously updating the thresholds PMF_THRE is provided, and is schematically illustrated in FIG. 3 . The unconditional PMF log(P(u=φ|Y 1 N )/P MAX ) are calculated for each received symbol, and these unconditional PMF are sorted from the largest value down to the smallest value, obtaining as many values SUPMF. Due to normalization with respect to largest unconditional PMF for every symbol, the first is 0, and for this reason it is not taken into consideration.

For each generic n-th largest unconditional PMF, the respective threshold PMF_THRE is preferably calculated as follows:

a moving average AUSPMF(n; i) of a number h of values of the sequence of n-th largest unconditional PMF of the incoming symbols is calculated for each received symbol i:

|SUPMF( n;i )−AUSPMF( n;i )|;

PMF_THRE( n;i )=AUSPMF( n;i )+ k·L 1( n;i );

wherein k is a parameter that may be fixed or programmed by the user according to the needs.

The above algorithm for determining the thresholds PMF_THRE(n; i) is relatively convenient because the parameter k may be determined substantially independently from the working conditions in which the symbols are received. Therefore, it may be possible to heuristically fix the parameter k only once, knowing in advance that the method will continue showing relatively good performance in terms of probability of missed detection or false detection of bursts even if the working conditions change.

The value L 1 ( n; i ) may be calculated differently from that shown above. Only for sake of example, it may be calculated using an L 2 norm:

or even a norm of a higher order, though the norm L 1 is preferred because it is simpler and faster to be calculated.

As an alternative to the above algorithm, it may be possible to process a limited subset of the sorted PMF of the generic i-th symbol, and not all of them to reduce the number of calculations, for example. In this case, the PMF of the limited subset may be considered for generating the sequence S(i).

FIG. 4 illustrates the performances of the method in terms of SFR vs. signal-to-noise ratio (SNR). In particular, the burst identification has a reduced affect on the performances in case the burst is not present. In fact a “random error” curve 44 , that is SFR in a random noise environment, and a “random error with identification” curve 45 , that is SFR in a random noise environment with burst identification switched on are almost superimposed.

Considering the case in which a 110 burst long is inserted, the curve 41 describes the performance of the genie-assisted receiver, meaning that the receiver knows where the burst is. The curve 42 describes the performance of the detection algorithm of the present embodiments. The loss is less than 0.1 dB. The line 43 represents the performance in case no burst detection is performed. The disruptive effect of the burst is evident.

The method described herein may be implemented by a computer program having software code or a computer readable medium storing software code adapted to make a computer execute the method. In particular, the computer program or computer readable medium may store software code that may make the computer identify, in real time, whether a received signal representing symbols, each of which may assume a value of a Galois Field, transmitted over a noisy channel, is corrupted by burst noise or not, according to the method described above.