ITMI20121353A1 - METHOD AND DEVICE FOR DEMODULAR WITH "SOFT" OUTPUTS QAM SIGNALS WITH GRAY MAPPING - Google Patents

Description

DESCRIPTION

The present invention relates to a method and a device for demodulating with "soft" outputs QAM signals with Gray mapping, according to the preamble of the corresponding independent claims.

Various methods are known for transmitting and receiving signals, the latter being analog or digital signals of radio power (television, telephone, satellite, etc.) or sent on copper supports, electrical lines or optical fibers. These signals may or may not be encoded and in the latter case, they are used to ensure greater reliability in terms of robustness against unwanted noise and / or interference or to ensure better performance (for example in terms of bit error rate) with equal noise and / or interfering power. If they are encoded, a channel encoder is introduced at the time of transmission (together with or before its modulation) and must be decoded at the time of its reception (after or together with a possible and appropriate demodulation).

Among the aforementioned transmission modes, the one that occurs with numerical amplitude modulation in quadrature, or QAM (Quadrature Amplitude Modulation), is known, where the signal is obtained by adding two signals (or carriers) modulated in phase and in quadrature. (in order to guarantee the orthogonality between them). This QAM signal can be represented through its own distribution on the complex plane, called constellation, which allows a concise and exact representation of the signal itself. All the possible states of each carrier can therefore be represented on the constellation for which the representation of the QAM signal in terms of a two-dimensional constellation (in the complex plane) is given by the set of points corresponding to the possible pairs of carrier amplitudes.

The bits of an information flow are inserted in the signal which must be suitably demodulated when it is received by a receiver.

In a QAM signal with Gray mapping, it is known that the bits of the information stream are represented in the constellation in such a way that adjacent symbols in the latter (i.e. the points representing, each, a number of bits equal to the logarithm to base 2 of the number of points in the constellation of the QAM signal) differ from each other by only one bit. The use of such a mapping thus ensures that a decision error, during the demodulation phase, towards a symbol adjacent to the one transmitted due to additive noises (for example a white Gaussian noise) corresponds to an error on a single bit , allowing, in the presence of channel coding, that the received signal is correctly decoded thus allowing to go back to the original transmitted signal.

In the transmission of the aforementioned signal, a channel encoder is preferably used which introduces redundancies in the transmitted signal to reduce noise in the received signal. The present invention considers that this encoder operates with a binary channel code (for example a convolutional code) which introduces redundancies with fixed memory and coding rate, regardless of the modulation order.

Solutions are known which operate a coded transmission of signals (electrical or radio) by means of a binary convolutional code at a fixed speed (rate) with a QAM modulator. In these solutions, in the signal transmitter the bits to be sent are grouped at the output of the encoder according to an order that depends on the preset one of the modulation. the latter must send to the decoder the reliability values relating to the bits (not the symbols) of each demodulated symbol (which includes several bits). The demodulator can have so-called "hard" outputs, that is the demodulator makes a decision (for a certain symbol of the constellation) in correspondence with each sample of the received signal, that is, it chooses (in a fixed way) the values of all the bits of the symbol ( more precisely, each bit is decided as 0 or as 1)

Alternatively, this modulator can have "soft" outputs, i.e. the demodulator generates reliability values, typically logarithmic likelihood ratios (or Logarithmic Likelihood Ratio, LLR), on each bit of the symbol that can be associated with the corresponding sample of the received signal, being the sample the value at the output of a sampler and an appropriate quantizer of the electrical signal at the input of the demodulator: the sign of the LLR is representative of the bit (0 or 1) and its modulus is representative of the reliability of the decision to favor of that bit. The greater the modulus of the LLR, the greater the reliability of the decision in favor of the bit corresponding to the sign of the LLR.

However, these known solutions require in the demodulator a digital processor which has a high capacity memory capable of allowing the necessary calculations to be performed to evaluate the LLRs to be associated with the bits contained in the received signals. Alternatively, the digital computer must be able to perform very advanced and complex calculations. In both cases, a demodulator with one or more digital processors of the type indicated above is very expensive to be able to perform all the calculations necessary for the "soft" output demodulation of the signal and its preparation for the eventual sending of the demodulated signal (meaning by signal demodulated the sequence of reliability values on the bits) to the decoder.

The purpose of the present invention is to offer a method and a corresponding device for demodulating with "soft" outputs a QAM signal with Gray mapping which are improved with respect to known solutions.

In particular, the object of the invention is to offer a method and a consequent device of the aforementioned type which can be implemented and manufactured at low costs with respect to known solutions.

Another object is to offer a device which can be manufactured with limited memory dimensions, so as to facilitate its use on fixed or mobile devices.

These and other objects which will be evident to the skilled in the art are achieved by a method and a device according to the appended claims.

For a better understanding of the present invention, the following drawings are attached purely by way of non-limiting example, in which:

Figure 1 represents a plurality of quantization sub-regions within a decision region of a symbol depicting an implementation step of the method according to the invention;

Figure 2 represents a graph of an optimized value for selecting the reliability value to be associated with a bit of a symbol obtained according to the invention; And

Figure 3 shows a block diagram of an operating mode of a device implementing the present invention.

According to the invention, the methodology for demodulating the QAM signals with Gray mapping of the state of the art is improved as the demodulator has "soft" outputs but generates approximate and quantized reliability values and that is starting from the approach to the approximate calculation of the LLR according to the 'approach proposed in A. J. Viterbi, "An Intuitive justification and a simplified implementation of thà ̈ MAP decoder for convolutional codes", IEEE Journal on Selected Areas in Communications, vol. 16, No. 2, pp. 260-264, Feb. 1998, and hereinafter referred to as the "Viterbi98 algorithm" (well known by the skilled in the art and therefore not further specified) which leads to the evaluation of approximate values of the LLRs defined as aLLR these aLLR are quantized by considering appropriate quantization thresholds . Performance evaluation with the use of aLLRq has shown that even using approximate reliability values (deriving from the above indicated appropriate quantization of aLLR) there is a completely negligible loss of performance if compared to what is obtained according to the known methodology that considers the exact calculation of the LLR or the calculation of the approximate values aLLR as proposed according to the Viterbi algorithm9 8.

According to an important characteristic of the invention, the aLLRq is quantized or the value of aLLR (calculated according to the approach proposed with the Viterbi algorithm98) is associated with a particular value of aLLRq depending on the region to which the aLLR belongs. This is obtained by considering a point or symbol s associated with a sample of the modulated signal (see figure 1). Consider also the neighboring symbols (si, s2, s3, s4) and the decision region R associated with the symbol s. For a specific bit of the latter symbol, the absolute value of the corresponding aLLR depends only on the distance, along the vertical or horizontal direction, of the edge of the border decision region with that of the adjacent symbol with corresponding bit opposite to that of the symbol of reference. On the basis of this observation it is possible to quantify the range of variability of the absolute value of the ALLR.

The method according to the invention provides for the choice of quantization levels. For example, the two-level quantization rule for calculating the ALLRq of the generic bit b of a symbol can be written as:

<'> c

2seO <| L (&) lâ ‰ ¤c ^ -

3C 1 /, C \

V â € "2 if £. (3⁄4)> c <= - 2â ‰ ¤ v <-2 (Va - 0 / l

where 2d is the distance between 2 adjacent symbols on the constellation and there is a parameter to be optimized.

Consequently, a two-value ALLR quantization corresponds to subdividing the decision region, with respect to the adjacent symbols, as shown in Figure 1.

In particular, within the "hard" decision region of the symbol s (in which the values of the bits corresponding to those of the symbol s are chosen), the nine quantization sub-regions {RÌ} <9> Ì = Indicated in figure 1. By indicating with bi the bit that changes between the symbol s and the symbol Sj (j = l ... r4), we have that according to the quantization sub-region where the (observable) sample r of the signal received, it is possible to automatically determine the quantized value of the aLLR, ie 1 'aLLRq. In general, the reliability value to be associated with each bit within each symbol is obtained from a table of quantized reliability values (aLLRq) like the one shown in Table 1.

TABLE 1

Table of quantized aLLRs (aLLRq) relating to the "hard" decision region of the symbol s in Figure

1.

R and L ibi) ^ <3> ^ L (b2) <quant> L (b3) <quant> L (b4) <quant>

j> 5

Ri c / 2 3c / 2 3c / 2 c / 2 3c / 2 R2c / 2 3c / 2 3c / 2 3c / 2 3c / 2 R3c / 2 c / 2 3c / 2 3c / 2 3c / 2 R43c / 2 c / 2 3c / 2 3c / 2 3c / 2

Rg 3c / 2 3c / 2 3c / 2 3c / 2 3c / 2 In particular, since for each adjacent symbol we have that 1 <1> aLLR of the reference bit (which changes between the considered symbol and the adjacent one) depends only on the distance from the corresponding edge of the R region, with simple geometric reasoning a two-level quantization of the ALLR corresponds to subdividing the R region into (2 + 1) = 9 sub-regions. Taking into account that bits that do not change when passing from the considered symbol to adjacent symbols will be associated with aLLR in very high modulus (according to the Viterbi98 algorithm, we arrive at Table 1: depending on the sub-region in which the observable falls (sample of the signal received), it is possible to emit the sequence of aLLRq, on the bits of the symbol, inserted in the corresponding row of the matrix.

In the general quantization rule introduced above, the parameter c appears: the optimization of the quantization rule is therefore reduced to the choice of an optimized value of c.

At this point, we proceed to optimize the parameter c in the presence of a binary channel coding. A possible approach to optimization is the following. For a given SNR value (i.e. the Signal-to-Noise Ratio), the bit error rate (BER) at the output of the decoder associated with the binary channel code used is evaluated, as a function of c: the optimized value, indicated as copt, corresponds to the minimum of the BER. The choice of the optimized value of c automatically determines the optimized subdivision of the R region into quantization sub-regions. In figure 2 the BER is shown as a function of c, for various SNR values, in the case of 64-QAM with convolutional encoding at rate 3/4 considering two quantization levels. From the results obtained it clearly emerges that, regardless of the SNR, Coptr2. Using this value for the two-level quantization of the ALLR results in optimized performance.

The proposed approach for two-level quantization can be extended in the case of multi-level quantization. In any case, the results obtained show that a two-level quantization introduces a very limited loss. In general, denoting by nquant-iev the number of quantization levels of the ALLR, the number of quantization sub-regions within a square decision region (region of an interior point of a constellation of a sufficiently high order) is (n ^ nt-iev + l) <2> (there are <n> quant-iev + 1 quantization "bands" in the horizontal direction and as many in the vertical direction).

From the above, it is possible to identify the memory occupation of the demodulator (or the digital processor present therein) operating according to the invention. In fact, assuming a square decision region for a symbol of a certain constellation, although reported previously in the case of quantization with nquant-ievlevels (nquant_iev + 1) <2> quantization sub-regions are required. For each quantization sub-region we need | -log2n ^ nt-ievi bits to indicate a certain quantization level, where fn) represents the smallest integer greater than or equal to the real number n. In the case of the M-air constellation, for each decision region we need a matrix of dimensions (n ^ ant ^ iev + 1) <2> x log2M; each data entered contains | -log2bit to index the aLLRq (ie the quantized value of aLLR). Therefore the total memory occupation required for an M-aria coded modulation is of the following order:

hIqLLR-i ~ M x (nquant-iev 1) x log2M x plog2nquanq-iev- | [bit]

This formula therefore depends on only two parameters: the modulation order M and the number of quantization levels n ^ nt-iev

In the case indicated above where n ^ ant ^ iev = 2 we have

MaLLR-q2 ~ 9 x M x log2M [bit].

From the above it can be seen the advantage, in terms of memory occupation, of an approach based on the use of aLLRq (as proposed) compared to non-quantized aLLR. It turns out that memory occupation is reduced by a percentage ranging from 75% (M = 4) to 80% (M = 4096). Since the performance losses are limited, this result makes the demodulation method with soft quantized outputs described above extremely advantageous, as it demonstrates that its strong point is the occupational efficiency in terms of memory.

In general, downstream of the reception of a signal sample, the operating steps of the demodulator with "soft" outputs with calculation of aLLRq include the following (as shown in Figure 3).

1) Hard decision on the most probable symbol (identification of the corresponding decision region) associated with the sample received (block 1 of figure 3).

2) Identification, within the decision region, of the appropriate quantization sub-region (block 2 of figure 3).

3) Reading of the matrix ((nquant-iev + l) <2> x log2M) corresponding to the indices (sequences of plog2 <n> quant-iev "| bit) of the aLLRq (block 3 of figure 3). 4) Reading of the actual values of aLLRq corresponding to the binary indices of an array of dimensions | -log2nquant-ievi (with real input data) (block 4 of figure 3).

These are the operating steps of a "soft" output demodulator that uses the aLLR optimized quantization rule proposed. It is, in fact, a demodulation method which requires in advance the creation of appropriate tables (generalizations of Table 1) relating to the emission of aLLRq.

A preferred form of the invention has been described. Obviously, in the light of the above description, the person skilled in the art will be able to identify other solutions similar to the one described, solutions which, even if operating according to different mathematical algorithms, can foresee the same implementation steps of the method reported in the following claims.

The device object of the patent and such as to implement the aforesaid method provides a demodulator having digital processor means or similar suitable to allow the identification of quantization sub-regions around each symbol belonging to a received modulated signal, memory means containing values corresponding tables of values of aLLRq associated with a generic signal, comparison means for comparing the reliability values or aLLRq for each sub-region identified and for generating a signal obtained on the basis of this comparison suitable to be decoded by a binary channel decoder .

Obviously also in this case, the demodulator device is connected to other elements and components of a more complex apparatus or system (for example an apparatus for receiving radio, electrical or optical digital signals) which includes at least one signal receiver and one user to which the decoded signals are sent (such as a screen, printer or computer memory, for example).

The invention therefore also relates to a system which comprises a device of the aforementioned type and according to the following claims.

Claims (10)

CLAIMS 1. A method for demodulating QAM digital signals with Gray mapping, said method comprising the reception of the QAM signal by a demodulator, the association of each sample of the signal received at the input of the demodulator to a symbol containing at least two signal bits, the symbols representing data associated with the signal being representable through a constellation, the generation of reliability values, approximating logarithmic likelihood ratios or LLR, for each bit of each symbol, corresponding to an observable sample, of the received signal and the generation of a signal demodulated, characterized by the fact that this demodulated signal is composed of approximated and quantized reliability values on the bit. 2. Method as per claim 1, characterized in that the following steps are envisaged: a) the choice of a number of quantization levels of the reliability value; b) the optimization of a quantization rule able to reduce the bit error rate or BER during the decoding following the demodulation; c) the definition of quantization sub-region within each decision region in which to evaluate the received QAM signal, said region being defined around each symbol of the QAM signal, the definition of the sub-region being carried out as a function of the level quantization chosen of the optimization of the quantization rule indicated in point b); d) the association of each quantization sub-region to corresponding reliability values to be assigned to the bits of each symbol of the received signal, said steps a) -d) being carried out before receiving the signal. 3. Method according to claim 2, characterized in that after receiving the signal it is expected: a) identifying a specific decision sub-region for each symbol associated with the received QAM signal; b) to read reliability values on the bits corresponding to the identified sub-region and to the selected quantization level; c) the generation of a coded signal consisting of optimized reliability values, said signal being sent to the demodulator output. 4. Method according to claim 3, characterized in that, after step c), said coded signal is sent to a binary decoder. 5. Method as per claim 2, characterized in that the choice of the number of sub-regions as a function of the quantization level is performed according to the following formula: number of sub-regions = (nquant-iev + l) <2> where is it nquant-iev = number of levels of quan tions chosen. 6. Method as per claim 2, characterized by the fact that the association of each quantization sub-region to a sequence of log2M bit reliability values leads to the creation of a table for each symbol present in the QAM signal in which they are reported the reliability values of the symbol bits depending on the quantization levels. 7. Device for carrying out the method according to claim 1, said device comprising a demodulator of a Gray-mapped QAM signal generated by a QAM modulator, the signal containing data represented, in a constellation, by symbols containing at least two bits, said demodulator being connected to a signal decoder, the demodulator being characterized in that it comprises: a) memory means containing tabulated and quantized values corresponding to decision reliability values relative to the bits of each symbol contained in the QAM signal, said values being functions of a preselected quantization level; b) comparison means for comparing the reliability values identified by a corresponding quantization sub-region of a decision region defined around each symbol of the QAM signal; c) means for generating a demodulated signal formed by optimized reliability values and for sending said demodulated signal to a binary channel decoder. 8. Device according to claim 7, characterized in that the channel decoder is a soft input decoder, a corresponding binary encoder being arranged upstream of the QAM modulator and receiving the signal to be transmitted. 9. Device according to claim 8, characterized in that the decoder is a convolutional decoder and the encoder is a convolutional encoder. 10. System for transmitting QAM signals with Gray mapping such as radio signals such as audio, video or data transmission signals, such as satellite, or the transmission of signals in a power supply line or a power transfer line such as a high voltage line, or the transmission of optical signals, said system comprising a device according to claim 7, in the system the Gray-mapped QAM signal being demodulated according to the method of claim 1.