GB2418802A - Phase sensitive diversity combining - Google Patents

Abstract

A method of receiving, via a transmission channel, a multicarrier spread spectrum diversity transmission using modulation onto a plurality of subcarriers, comprising combining plural said subcarriers using a diversity combination technique which is sensitive only to the relative phase shifts of said subcarriers due to the channel, and not to the relative gains thereon. The modulation may be M-PSK and the diversity combining may use Selective Combining (SC) or Equal Gain Combining (EGC).

Description

24 1 8802 Multi Carrier communications This invention relates to

modulating/demodulating and decoding Multi Carrier (MC) signals. More particularly, this invention relates to decoding using the maximum a Posteriori (MAP) decoding technique. Particularly but not exclusively, the invention is used for decoding turbo encoded signals, and particularly but not exclusively in mobile telecommunication systems.

Multi-carrier systems are today commonly used, with the IEEE standardization of for example IEEE 802.1 la and g or IEEE 802.16a, or the ETSI standardization of the Digital Audio Broadcasting (DAB) or the Digital Video Broadcasting (DVB-T) systems.

Multi-carrier modulation is a diversity data transmission technique where several subcarriers are used to transport the user data signal.

The enhancement of robustness to multi-path, flexibility and reduced complexity compared to other powerful techniques such as code division multiple access (CDMA) systems, has led to the introduction of new radio air interfaces, such as multicarrier CDMA systems (MC-COMA). Such systems are for example described in A. Chouly et al, "Orthogonal multicarrier techniques applied to direct sequence spread spectrum CDMA systems", GLOBECOM '93 or in N. Yee et al., "Multicarrier CDMA in Indoor Wireless Radio Networks", Proceedings PIMRC'93, pp. 126-133. MC systems have therefore been considered as a good candidate for Next Generation Wireless Enablers, see for example WWRF, Book of Vision, Rask 4.7, "Multi- Carrier Based Air interface", pp232, http://www.wireless-world-research. org/, MAGNET, http://www.telecom.ece.ntua.gr/magnet/ or WINNER, http://www.ist-winner.org/ Sixth Framework IST Project.

Next Generation air interface proposals involve thousands of subcarriers, and suggest user multiplexing by using Orthogonal Frequency Division Multiple Access systems (OFDMA), which has been proven to reduce intercell interference and significant packet transmission flexibility.

S. Kaiser and K. Fazel proposed a hybrid scheme in the paper "A flexible Spread-Spectrum Multi-Carrier Multiple-Access System for Multi-Media Applications, in Proc. IEEE PIMRC'97, pp.100-104, namely MC-SS.

In MC-SS systems, frequency spreading of the user data is applied similarly to MC-CDMA systems. However, the orthogonality among the users is here obtained by multiplexing on distinct subcarriers. In this way the advantage of the frequency diversity offered by the spreading is combined with the flexibility of user multiplexing offered by OFDMA systems, see for example the paper by S. Pietrzyk and G.J.M Janssen, "Multiuser Subcarrier Allocation for QoS Provision in the OFDMA Systems", IEEE VTC 2002, Vol.2, pplO77-1081.

Transmission systems can be further improved by the use of channel coding. Channel coding is a method of adding redundancy to information so that it can be transmitted over a degrading (fading and/or noisy) channel, and subsequently be checked and corrected for errors that occurred in transmission. Channel coding is especially beneficial for wireless and multimedia applications.

Turbo codes are widely used in challenging communications environments such as mobile telephony because they offer a transmission performance which approaches the Shannon limit in noisy conditions. In turbo coding, two recursive systematic convolutional (RSC) coders are provided (usually) in parallel, receiving the same information bits but in different orders. They are described in Berrou C., Glavieux A. & Thitimajshima P., 'Near Shannon Limit Error-correcting Coding and Decoding: Turbo Codes (1)', IEEE Int. Conf. On Comm., May 1993.

The output of the turbo coder is the systematic information (the original bit stream) which forms the input to one of the RSC coders, together with the parity outputs of each of the RSC coders. This corresponds essentially to two stochastic processes running on the same input data. At the receiver end, a turbo decoder can operate by independently estimating each of the two processes with a decoder, then refining the estimates by iteratively sharing information between the two decoders. In other words, the output of one decoder can be used as a-priori information by the other decoder.

Each decoder therefore produces a soft output indicating the likelihood that the input bit was a one, referred to as a "log likelihood ratio" (LLR).

Each decoder is arranged to accept the received bit stream (comprising systematic imparity information) together with a-priori information (in the form of an LLR or data derived from it) from the other decoder, and generate an output (which can itself be fed to the other decoder).

The calculations which are performed during each iteration are described in, for example, "Implementation of a 3GSP Turbo Decoder on a Programmable DSP Core", James G. Harrison, 3DSP Corporation White Paper, presented at the Communications Design Conference, San Jose, California, October 2, 2001 (www.3dsp.com/pdf/3dspturbowhitepaper.pdf).

The MAP algorithm for turbo decoding MC signals results in a very high complexity receiver which is therefore expensive and power-hungry, both of which make it difficult to incorporate into mobile terminals which should be cheap and have a long battery life, particularly because of the computation of a large number of "branch metrics". The present invention therefore aims to provide a transmission scheme allowing a lower complexity receiver, and a receiver for use with such a scheme.

The invention rests on the realisation that, for a small penalty in performance, a drastic simplification of the decoder can be made if it ignores state information and only makes use of phase information, using for example Equal Gain Combining (EGC) or Selective Combining (SC) as the diversity reception technique. Although simplification of the decoder is the main benefit, embodiments may also require fewer pilot signals than in the prior art, and hence have more usable bandwidth, somewhat offsetting the abovementioned small penalty in accuracy.

A general expression of the mathematics underlying the simplification of the decoder would in be intractable, but an approximation has been found which provides good performance. Since the system uses only estimated phase and not estimated channel state for each channel, M-ary Phase Shift Keying (M-PSK) can be used as the modulation technique at the transmitter.

Preferably, the invention is utilised in a mobile telecommunication system; it may be either at the user terminal end or at the system end, but is particularly useful in the former case where the computation hardware available is more constrained.

Other aspects, embodiments and preferred features will be apparent from the following description and claims. Embodiments of the invention will now be illustrated, by way of example only, with reference to the accompanying drawings in which: Figure 1 is a block diagram showing schematically the elements of a communications system according to an embodiment of the invention; Figure 2 is a block diagram showing schematically the elements of a receiver within a user equipment mobile terminal forming part of Figure 1; Figure 3 is a block diagram showing schematically the elements of a transmitter within a base station forming part of Figure 1; Figure 4 is a diagram schematically illustrating a Parallel Concatenated Convolutional Coder (Turbo Coder); Figure 5 is a diagram schematically illustrating an iterative Turbo Decoder (MAP decoder); Figure 6 is a flow diagram schematically illustrating the process of iterative Turbo-Decoding; Figure 7 is a plot of simulated and theoretically calculated results for mean signal to noise ratio output (SNR) against number of diversity branches; Figure 8a is a plot of conditional probability density (marginalised with respect to fading) in low SNR conditions for minus 1; Figure 8b is a plot of conditional probability density (marginalised with respect to fading) in low SNR conditions for plus 1; Figure 9a is a plot of conditional probability density (marginalised with respect to fading) in low SNR conditions for minus 1; and Figure 9b is a plot of conditional probability density (marginalised with respect to fading) in low SNR conditions for plus 1.

Overview of System Referring to Figure 1, a communication system such as a mobile telephony system comprises a plurality of user equipment (UK) such as mobile terminals 300a, 300b, in radio communication with a base station 100, provided within a cell, and having a fixed link connection to a backbone network (such as an IP network) via a switch computer 200. The IP backbone network will not further be discussed since it is of conventional type.

Receiver (at mobile terminal) In both the base station and the terminals, in order to receive data transmitted over the communications link, the received data are first demodulated, then processed in error control decoders using the redundancy to correct any errors occurred in transmission, such that the data are then S available at the receiver's output port (for example for display or audio reproduction). The following description of a receiver is in relation to the mobile terminal but is also applicable to the network side (although it is less important to achieve low complexity at the relatively few, relatively expensive base stations) Referring to Figure 2, the mobile terminal 300 comprises a control unit 302 such as a microcontroller, a power supply 304 such as a rechargeable battery, and a user interface 306 such as a touch screen. A downlink receiver antenna 312 receives the OFDM radio signal transmitted by the base station and passes it to a downlink receiver 314 at which it is amplified and down converted to base band and supplied to an OFDM demodulator 315 (e.g. performing an FFT). Channels from the OFDM demodulator 315, under control of a channel selector 310, pass to a demultiplexer 322 which separates out the pilot signals. These are routed to a phase correction unit 320, at which the phase is measured, and phase errors (introduced by the channel) for all other subcarriers are calculated by interpolation from the measured phase values.

Channel decoding is performed by a channel decoder 316, with which the present invention is principally concerned as discussed further below. The decoded data is then passed to an output processing unit (not shown) in which higher layer processing is performed, for example to remove source coding, reassemble packets and so on, and the output data from the output processing unit 318 is supplied for use, at a data port either to the user interface 306 for display, to the control unit 302, or to an output port assumption by an external device.

A transmitter is also provided within the mobile terminal, for sending uplink signals, but further description is not necessary for the understanding of the present embodiment.

It will be clear to the skilled person that, other than the modulation and RF components, the blocks making up the embodiment at the base station and the mobile terminals comprise suitably programmed digital signal processor devices (DSPs, comprising a digital signal processor with associated RAM flash memory, storing a control program and scratch calculations performed during the operation of the terminal) or ASICs executing signal processing.

Separate dedicated hardware devices may be used for the OFDM operations (specifically the Inverse Fast Fourier Transform or IFFT used in the transmitter components to map the subcarriers into a composite time domain signal for RF modulation) and the Fast Fourier Transform or FFT used in the receiver components to map the time-domain signal back into component subcarriers).

Referring to Figure 3, on the transmitter side of the base station 100, the data signals intended for terminals 300 are received from the network computer 200 and multiplexed, together with pilot signals and control signals from the control unit 120, onto selected OFDM subcarriers. A carrier allocator 110 determines which subcarriers will be used to communicate with each mobile terminal 300, and controls the multiplexer accordingly.

The binary data are encoded and modulated in channel coder 12, by Turbo coding the data bits 126, interleaving 112, and applying M-ary Phase Shift Keying (M-PSK) 1 15.

The data stream is subsequently fed to a serial to parallel converter block 114 and is thereby multiplexed into N parallel data streams, i.e. the N Spreading Bands, constituted of L subcarriers each.

In diversity block 116 each data stream is spread over L subcarriers using spreading chip codes C1 to CL, i.e. by duplicating the data stream L times and multiplying by the chip elements Cl to CL (one chosen spreading sequence) at multiplier bank 1 17.

Specific frequency mapping is applied in block 118 and pilot tones (i.e. symbols having a predetermined phase) are supplied on selected subcarriers spaced apart in frequency by pilot tone unit 119. In block 122 an inverse Fast Fourier transformation is carried out and the data are fed into parallel to serial converter, wherein the data are added.

The OFDM signal is then RF-up converted and amplified and supplied to a downlink transmit antenna (not shown), and sent through the channel for reception by the mobile terminals 300.

Uplink channels are also provided, in this embodiment, for receiving uplink data, but further description is not necessary for the understanding of the present embodiment.

Turbo-Coding & De-coding Referring to Figure 4, the channel coder 326 of the user equipment 300, and the equivalent channel coder of the base station 100, comprise a turbo coder consisting of an input port 3262 which accepts an input bit stream; a pair of equivalent recursive systematic convolutional (RSC) encoders 3264a, 3264b fed in parallel from the input port 3262; an interleaver3266 in the path between the input port and one of the RSC encoders 3264b; a puncturing unit 3268 for varying the coding rate by selectively removing bits from the output stream; and an output port 3270 comprising a multiplexer for multiplexing into a single bit stream the input bit stream from the input port3262, and the punctured output streams of the encoders 3264a, 3264b via the puncturing unit 3268.

This corresponds to the parallel concatenation of two recursive systematic convolutional (RSC) encoders. These encoders are separated by one interleaving module fir. As these encoders are processing the same input bits stream xs, this scheme is referred to as "parallel concatenated convolutional codes" (PCCC).

Each RSC component is composed of two generator polynomials, namely feedback and feedforward polynomials, respectively g' (D) and g2 (D). Thus, two parity check sequences are generated, namely A, and x2P.

The turbo-decoding process which takes places in the receivers at the base station and the terminals uses the basic iterative scheme. Again two component decoders, called Soft Input Soft Output (SISO) decoders Do and D2, are serially concatenated via an interleaving process. The major key feature of this iterative process, is that the extrinsic information delivered by one SISO decoder, feeds the following SISO decoder as a priori information for decoding information bits. The feedback loop iteratively improves the performance of this scheme.

Referring to Figure 5, the channel decoder316 comprises an input port 3162 at which the received modulated samples (represented as multibit data) are demultiplexed into a data stream and two separate parity streams.

Each parity stream is fed in parallel to a respective soft input soft output (SISO) decoder 3164a, 3164b. An interleaver 3166 of depth equivalent to the interleaver3266 decoder lies in the path from the input port3162 to the second decoder 3164b.

Each of the coders3164 has four ports. A first input port31642 (shown only on the first decoder3164a for clarity) accepts the observation samples from the input port 3162 and a second 31644 accepts a- priori information from an extrinsic information output port 31646 of the other decoder. Finally, an a posterior) output port supplies decoded output bits from each decoder3164 to a combiner and thresholder3168 at which the decoded bit-stream is made available as a series of binary bits.

In the output path from the second decoder3164b to the output port 3168 is an output deinterleaver 3169.

Interconnecting the output port 31646 of the first decoder and the a priori input port31644 of the second decoder3164b is an interleaver3171.

Conversely, interconnecting the output port 31646 of the second decoder 3164b with the a-priori input port 31644 of the first decoder 3164a is a deinterleaver 3173.

Referring to Figure 6, an overview of the operation of the decoder of Figure 5 will be given.

Step 1002, a new demodulated sample (in the form of a 32 bit value from the demodulator) is received at the input port.

In step 1006, each decoder computes and stores the branch metrics (often referred to as gamma values) for all trellis branches. The skilled person is aware that a trellis diagram is a representation of the different states applied by the modulator over a succession of bit periods (equal in length to the length of the interleaver plus a predetermined number of periods for flushing the encoder back to an initial state). The decoder attempts to trace the path through the trellis taken by the signal by calculating the state transition probability p(ylx).

The branch metrics (gamma values) are calculated based on the distance between the "hard" actual coder values and the "soft" received values from the demodulator, taking into account the channel noise variance, using a relationship described in greater detail below, multiplied by the a-priori input from the other decoder. Calculation of the branch metrics forms a substantial computational task.

After computing and storing the branch metrics, in step 1008, each decoder performs a forward recursion on the trellis by computing forward (alpha) probabilities for each node on the trellis. The alpha values are the sum of the previous alpha multiplied by the branch metric along each branch from the two previous nodes to the node in question.

In step 1010, each decoder performs backward recursion to calculate backwards (beta) probabilities for each node, similarly to the forward recursion but starting at the end of the trellis and going backward.

In step 1012, the log likelihood ratio (LLR) for each time t is calculated as the sum of the products of the alphas, betas and gammas at that time for each branch associated with a one value in the encoder, divided by the sum of the products of the alphas, betas and gammas for each branch associated with a zero.

In step 1016, the decoders determine whether the maximum number of iterations (for example, four iterations), Niter, has been reached. If not, then in step 1018 each decoder calculates extrinsic information to be fed from each coder to the other is calculated as the difference between the LLR and the input probability estimate and in step 1020 outputs the extrinsic information at its output port.

In step 1022, each of the two decoders accepts a new a-priori value from the other decoder.

When all iterations have been complete (step 1016) the output device thresholds the log likelihood ratio outputs of the two decoders to form the decoded output bit. If the bit rate is sufficiently slow, the test 1016 may determine whether the decoded outputs have converged, rather than applying a limited number of iterations.

The MAP algorithm, given below, represents the optimal soft-output processing: a, (I) y, (I,1) (1) (c, ) = log (, I')eB' (2) a,-,(I') r, (1',1) ,(1) (I I')eB This computation is performed by the forward-backward algorithm in the

prior art.

In step 1018, the extrinsic information to be fed from each coder to the other is calculated as the difference between the LLR and the input probability estimate.

For a Rayleigh channel, Y (Yk Yk ak ak, m, m) x P(Yk |Uk = i, Ok) (3) and the conditional probability is +w P(Yk |Uk i) = Jp(a) p()J' |uk a) da 0 (4) This integral is intractable. However, we have realised that it can be approximated as a Gaussian around the term of interest.

For Selective Combining, in which the single best signal is selected (k is the number of the branch up to the spreading factor L), Yk = Xk MaX[ak '---'at)]+ nk (5) If we denote: ak = Max [ak( ),.. ., ak) (6) the conditional probability (i.e. the MAP branch metric) is given by P(yklxk ak akL))= I (Yk -X. a)2 1 1 0 (7) Marginalising (minimizing) this with respect to fading gives += P Ark |Xk) JP (Ok |Xk ak) p (ak) dak o (8) Now, the MAP branch metric can be re-written as a Gaussian relation: ply IX)= N( () M (A No)) (9) where the variance of the channel amplitude is given by: cry = El(a (a)) ]= E[a2]- (a)2 (10) and the mean is given by: +a' (a) = E(a) = ia p(a) da o (1 1) Now, some manipulation of the above leads to a new expression for the mean as follows: (a) 2 ( (k) (12) and: E[a2]= ia p(a) da=(-lk Ink) k (13) which can be substituted back into Equation (9) to give a result that depends on the measured Noise (No) per channel, and mean SNR per subcarrier (, and not on channel amplitude or fading, so that they do not need to be measured.

It will be seen that Equation (9) has two regimes; a low noise regime where the first term of the Max() expression dictates the behaviour of the expression based on fading variance, and a high noise regime where the second term of the Max() expression dictates the behaviour of the expression based on the level of noise.

Results of simulations Figure 7 shows a plot of theoretical (dashed line) and simulated according to the embodiment (circled points) mean output SNR against number of diversity branches L. This is a measure of the accuracy of Equation 12 above. It is clear that there is a good match.

Figures 8a and 8b measure the accuracy of Equation (9) above for low SNR conditions. They show the conditional probability p(ylx) marginalised with respect to fading for simulated real data for 20,000 symbols, using L=4, . In this case, the SNR in dB is 6dB so No/2) is 0.5012. The theoretical mean is 1.3885.

Figure 8a shows the resulting density for -1. The mean is ( 1)*1.3766, and the variance is 0.5345. Figure 8a shows the resulting density for +1. The mean is (+1)*1.3903, and the variance is 0.5277. These figures are very close to the predicted figure of 0.5012.

Figures 9a and 9b measure the accuracy of Equation (9) above for low SNR conditions. They show the conditional probability p(ylx) marginalised with respect to fading for simulated real data for 20,000 symbols, using L=4, for high SNRs. In this case, the calculated variance is 0.3942.

Figure 9a shows the resulting density for-1. The variance is 0.3936.

Figure 9a shows the resulting density for +1. The variance is 0.3976. These figures are very close to the predicted figure.

Thus, in each case, there is good agreement between the simulations and the theoretical model based on the above Equation (9).

Other Embodiments and variants It will be clear that the above described embodiments are examples only, and that many other embodiments are possible.

For example, although selective combining (SC) is disclosed, Equal Gain Combining, which likewise requires no knowledge of signal amplitude but only phase, could equally be employed. Combinations of SC and EGC, or other such techniques, are equally applicable. Other techniques than turbo coding could be used for the channel coding, and techniques other than PSK could be used for the modulation. The invention has been disclosed for mobile communications (two way communications) but is equally applicable to broadcast (unidirectional) communications.

The invention extends to any and all such variants and modifications which would be obvious to the skilled person from the forgoing. For the avoidance of doubt, protection is hereby sought for any and all new subject matter and combinations thereof herein disclosed.

Claims (10)

1. Claims 1. A method of receiving, via a transmission channel, a

   multicarrier spread spectrum diversity transmission using modulation onto a plurality of subcarriers, comprising combining plural said subcarriers using a diversity combination technique which is sensitive only to the relative phase shifts of said subcarriers due to the channel, and not to the relative gains thereon.
2. 2. A method according to claim 1 in which said diversity combination technique comprises selective combining.
3. 3. A method according to claim 1 in which said signal is turbo coded, and further comprising applying a decoding technique.
4. 4. A method according to any preceding claim comprising applying Multiple A Priori (MAP) decoding.
5. 5. A method according to any preceding claim, in which the signal is phase shift keyed.
6. 6. A method according to any preceding claim, in which the signal comprises a plurality of pilot signals, and further comprising estimating said phase error due to said channel by interpolation from the phase measured on said pilot signals.
7. 7. A method according to claim 4, comprising calculating the MAP branch metric based on the mean amplitude on subcarriers, the variance (or square thereof) of fading on subcarriers, and the measured noise level.
8. 8. A MAP decoder performing the method of claim 4.
9. 9. A radio receiver comprising a MAP decoder according to claim 8.
10. 10. A mobile terminal comprising a radio receiver according to claim 9.