Channel Coding and Link Adaptation

Transcription

1 Seminar Ausgewählte Kapitel der Nachrichtentechnik, WS 2009/2010 LTE: Der Mobilfunk der Zukunft Channel Coding and Link Adaptation Shahram Zarei 16. December 2009 Abstract In this work channel coding and link adaptation in LTE are considered, which are important issues in modern digital communication systems. With channel coding, errors caused by distortion during transmission are detected and/or corrected. In LTE both convolutional and Turbo codes are used. The structure of convolutional codes in LTE is presented here. Turbo codes and internal contentionfree interleaver, which is an important part of the Turbo encoder are also topics of this work. The concept of the circular buffer, which is used in the rate matching module and HARQ is discussed, too. Another key feature used in LTE is the link adaptation. Link adaptation makes the efficient use of the channel capacity possible, matching the transmission parameters, modulation scheme and coding rate to the channel conditions. 1 Introduction Channel coding is one of the most important aspects in digital communication systems, which can be considered as the main difference between analog and digital systems making error detection and correction possible. Error correction exists in two main forms: ARQ (Automatic Repeat Request) and FEC (Forward Error Correction). With ARQ the receiver requests retransmission of data packets, if errors are detected, using some error detection mechanism. In FEC some redundancy bits are added to the data bits, which is done either blockwise (so-called block coding) or convolutional, where the coded bit depends not only on the current data bit but also on the previous bits. In LTE both block codes and convolutional codes are used. There is also an enhanced coding technique used in LTE, called Turbo code, which has performances within a few tenth of a db from the Shannons limit.

2 2 Shahram Zarei Another feature of LTE, which is considered here, is link adaptation. Link adaptation is referred to a mechanism matching automatically transmission parameters to the channel. As an example for older systems of link adaptation the early versions of UMTS (Universal Mobile Telecommunication System) can be mentioned, where fast closed-loop power control used to support an almost constant data rate. In UMTS the UE (User Equipment) transmitter adjusts its output power in accordance with one or more Transmit Power Control (TPC) commands received in the downlink, in order to keep the received uplink Signal-to-Interference Ratio (SIR) at a given SIR target. In HSPA (High Speed Packet Service Access) and LTE the transmitted information data rate is adjusted dynamically to use the channel capacity efficiently. 2 Link adaptation and feedback computation 2.1 Link adaptation in LTE In LTE, link adaptation is based on the Adaptive Modulation and Coding (AMC). AMC can adapt modulation scheme and code rate in the following way: Modulation scheme: if the SINR (Signal-to-Interference plus Noise Ratio) is sufficiently high, higher-order modulation schemes with higher spectral efficiency (hence with higher bit rates) like 64QAM are used. In the case of poor SINR a lower-order modulation scheme like QPSK, which is more robust against transmission errors but has a lower spectral efficiency, is used. Code rate: for a given modulation scheme, an appropriate code rate can be chosen depending on the channel quality. The better the channel quality, the higher the code rate is used and of course the higher the data rate. In LTE for data channels a Turbo encoder with a mother code rate of 1/3 is used. There is a Rate Matching (RM) module following the Turbo encoder, which makes it possible to get other code rates, if desired. Increasing and decreasing the code rate is done via puncturing and repetition, respectively. Both, puncturing and repetition are integrated in the Rate Matching module. In Fig. 1 the whole signal generation chain of the LTEs physical layer with Turbo coding and modulation modules can be seen, which are parts of the link adaptation system. 2.2 CQI feedback in LTE In LTE downlink, the quality of channel is measured in the UE and sent to the enodeb in the form of so-called CQIs (Channel Quality Indicator). The quality of the measured signal depends not only on the channel, the noise and the interference level but also on the quality of the receiver, e.g. on the noise figure of the analog front end and performance of the digital

3 Channel Coding and Link adaptation 3 Figure 1: Signal generation chain in LTE signal processing modules. That means a receiver with better front end or more powerfull signal processing algorithms delivers a higher CQI. The signal quality measurments are done using reference symbols. Depending on the SNR (Signal-to-Noise Ratio) a combination of modulation scheme and code rate is selected to ensure that the BLER (Block Error Rate) is less than 0.1. This can be seen in Fig. 2 Figure 2: BLER versus SNR for various combinations of modulation scheme and code rate In Fig. 3 a list of 4-bit CQIs corresponding to the 16 possible combinations of modulation scheme and code rate is shown. As can be seen in this table, CQI = 1 refers to the most robust transmission parameters i.e. QPSK as modulation scheme and the lowest code rate of 0.076, which is selected for the worst channel quality. With increasing channel quality, higher order modulation schemes and higher code rates can be selected. The highest order of modulation and highest code rate, which can be selected are 64QAM and 0.93 respectively and correspond to a CQI value of 15. In the above discussions it is assumed, that the considered channel is a slow fading channel. In other words, between two CQI measurments, channels behaviour does not change or the coherent time of the channel is at least as long as the CQI measurment period.

4 4 Shahram Zarei Figure 3: 4-bit CQI table Here, only the link adaptation for downlink is discussed. The link adaptation in uplink is very similar. The difference is, that enodeb estimates channel quality by evaluating the channel using SRSs (Sounding Reference Signals). In the LTE physical layer, resources are managed with the so-called RM Modules (Resource Management), which assign incoming data blocks to resource blocks. One resource block consists of 12 sub-carriers and one time slot. The resource management in LTE can be seen in Fig. 4. CQI values are used also to select the optimum resource block i.e. the optimum sub-carrier and the optimum time slot. Figure 4: Two dimensional resource management in LTE There are two kinds of CQI reporting: periodic and aperiodic, where the PUCCH (Physical Uplink Control CHannel) is used for periodic CQI reporting only and PUSCH (Physical Uplink Shared CHannel) for aperiodic CQI reporting. Periodic CQIs are reported by the UE in periodic time intervals. If the enodeb wishes channel quality information at a specific time, aperiodic CQIs are triggered. In order to define the frequency granularity of the CQI, the whole system bandwidth is divided into N sub-bands, each consisting of k contiguous Physical Resource Blocks (PRBs). The

5 Channel Coding and Link adaptation 5 number of sub-bands is given by N = NRB/k DL and determines the frequency granularity of the CQI reporting, where NRB DL is the number of resource blocks (RB) in the whole system bandwidth (DL stands for Downlink). 2.3 Periodic CQI reporting The periodic reporting of the CQIs is done over the PUCCH. Periodic CQI can be either wideband or UE-selected sub-band feedback for all downlink transmission modes. The type of CQI is decided by the enodeb. In the wideband mode, one CQI value is measured in the whole system bandwidth and sent to the enodeb. In the UE-selected sub-band feedback the total number of sub-bands N in the whole system bandwidth is divided into j fractions called bandwidth parts. In each bandwidth part a particular sub-band is selected and the measured channel quality in this sub-band with its position in the bandwidth part is sent to enodeb. In Fig. 5 sub-band size (k) and bandwidth parts (J) versus downlink system bandwidth N DL RB can be seen. Figure 5: Number of resource blocks in the whole system bandwidth (N DL RB), number of resource blocks in a sub-band (k) and bandwidth parts (J) in periodic CQI using UE-selected sub-bands 2.4 Aperiodic CQI reporting Normally periodic CQIs are used but if enodeb needs channel quality information at times rather than time raster of the periodic CQI, it can also wish aperiodic transmission of the CQIs by the UE. Loss of synchronization or handover situations are also cases, where aperiodic CQIs are used. Aperiodic CQI reporting is done over the PUSCH and requested by the enodeb by setting a CQI request bit on the Physical Downlink Control CHannel (PDCCH). The type of CQI is set by the enodeb and can be one of the following modes: Wideband feedback: in this mode as in the periodic reporting, the UE reports one CQI value for the whole system bandwidth. enodeb-configured sub-band feedback: there are two kinds of CQI reported in this mode, one for the whole system bandwidth and one for the sub-bands. In the calculation of subband CQIs, it is assumed that transmission takes place only in the relevant sub-band.

6 6 Shahram Zarei UE-selected sub-band feedback: as in the enodeb-configured mode, two types of CQIs are used, one wideband CQI value for the whole system bandwidth and one for reporting the average measured CQI in M selected sub-bands each of the size k. The UE decides, which sub-bands are selected. The UE sends also the position of the M selected sub-bands. Number of resource blocks in the whole system bandwidth, number of resource blocks in a sub-band (k) and number of selected sub-bands (M) by the UE for the UE selected sub-band feedback CQI reports can be seen in Fig. 6. Figure 6: Parameters of the aperiodic CQI report for UE-selected sub-bands feedback 3 Channel coding in LTE One of the most important issues in digital communication systems is error correction. Error correction can be considered in two main categories: ARQ (Automatic Repeat Request) and FEC (Forward Error Correction). With ARQ the receiver requests retransmission of the data packets, if errors are detected and this will be done, until the received packets are error-free or a maximum number of retransmissions is reached. For FEC redundancy bits are added to data bits, making error correction possible. FEC has two main types: block codes and convolutional codes. In block codes, the input data blocks, which can be considered as vectors are multiplied by a generator matrix, generating a code word vector. In contrast to block codes, convolutional codes have a similar structure like FIR (Finite Impulse Response) filters and operate bitwise. Convolutional codes have memory, which means the coded output bit depends not only on the current bit but also on the m previous bits, where m is the number of registers in the convolutional encoder. In LTE both block codes and convolutional codes are used. CRC (Cyclic Redundancy Check) which is a cyclic linear block code is used for HARQ as an error correction technique. Control channels and data channels in LTE use convolutional and Turbo codes, respectively, where the later is an enhanced development of convolutional codes achieving near-shannon performance. 3.1 Convolutional encoder in LTE A convolutional encoder C(k, n, m) consists of a shift register with m stages, whose outputs are added (XOR) together to build the output bit. Per k bit input information bits, n output bits

7 Channel Coding and Link adaptation 7 are produced from the encoder. The rate of the code is defined as: R = k/n. The structure of LTEs convolutional encoder with k = 1, n = 3, m = 6 and resulting code rate of 1/3 can be seen in Fig. 7. The n outputs in the convolutional encoder are calculated via n generator sequences: G = [g 0,..., g n 1 ]. The generator sequences used in LTE are: g 0 = [1 3 3](oct), g 1 = [1 7 1](oct), g 2 = [1 6 5](oct) Figure 7: LTEs convolutional encoder Decoding of convolutional codes In order to minimize the bit-error probability, in receiver a sequence should be selected, which maximizes the so-called MAP (Maximum A-Posteriori) probability: ˆx = arg max P (x y) (1) x Where x denotes the transmitted sequence and y the received sequence. If the codewords are equiprobable, the MAP criterion is equivalent to Maximum Likelihood (ML) criterion: ˆx = arg max P (y x) (2) x Instead of maximizing P (y x), log P (y x) can be maximized, because log(.) is a monotonically increasing function, which leads to the loglikelihood function for a memoryless channel: For an AWGN channel we have: log P (y x) = L+m 1 i=0 n 1 j=0 log P (y i,j x i,j ) (3) P (y i,j x i,j ) N( E b x i,j, N 0 ), (4) where E b is the transmitted bit energy and N 0 the variance of the white gaussian noise. Hence we get: log P (y x) y i E b x i 2 (5) In order to maximize equation (5), in receiver a codeword should be selected, whose euclidean distance to the received sequence is minimum. But examining all of the possible candidates for the input sequence is very costly and has exponential complexity.

8 8 Shahram Zarei A very efficient method to decode convolutionally coded sequences is the viterbi algorithm, which has much less computational complexity using the trellis diagram. The trellis diagram is a state diagram, where the time progress is also presented among the x-axis as can be seen in Fig. 8. In the example considered here, there are two registers in the convolutional encoder and therefore four states in the trellis diagram. Starting at an initial state and depending on the input bit, the next state is determined. Therfore, each input sequence can be represented with a path in the trellis, if the initial state is known. In order to decode a convolutionally encoded sequence the initial and final states should be known to the receiver. There are several methods to do this, two of them are: Insertion of termination bits (or tail bits) at the end of the information block to reset all shift registers to zero at the end of the encoding process. Tail-biting, where the initial and final states of the encoder are identical. Tail-biting is done by initializing the shift register contents with the last m information bits in the input block. The advantages of this method is the uniform protection of the information bits and that there is no rate loss Viterbi algorithm The Viterbi algorithm searches the best path in the trellis diagram with known initial and final states, which matches to the input sequence. Each path consists of branches, which connects two states and have certain cost (metric). The overal cost of a path is the accumulated metric of its branches. The aim in the viterbi algorithm is to find the path with the minimum cost. Branch metric at the i th trellis step (the cost of choosing a branch at the trellis step i) is defined as: M(y i x i ) = n 1 j=0 log P (y i,j x i,j ) (6) At the l th transition in the trellis, a partial path metric can be derived, which is the accumulated metric of branches between the initial state and the considering lth transition, following a specific path inside the trellis. The partial path metric can be defined as: M l s(y x) = l i=0 M(y i x i ) = M l 1 s (y x) + M(y l x l ) (7) Viterbi algorithm finds the best path through the trellis, by comparing all of the arriving edges in the state s and selecting a survivor edge with the best metric. It can be shown, the best edge is the one with the smallest hamming distance between the received sequence and the state sequence. At each step the survivor partial paths with the smallest accumulated metrics are selected and finally the best path between the initial and final state with smallest metric is selected, which corresponds to a specific input bit-sequence.

9 Channel Coding and Link adaptation Turbo coding in LTE Figure 8: Trellis diagram of a convolutional encoder Turbo codes were first introduced by Berrou, Glavieux and Thitimajshima in 1993 and have Performances within a few tenth of a db from the Shannon limit. The Turbo encoder in LTE is a systematic parallel concatenated convolutional code (PCCC) with two 8-state constituent encoders (the same as in UMTS) and one Turbo code contetntion free internal interleaver (different from UMTS). Where the encoders are based on RSC (Recursive Systematic Convolutional) codes and their generator polynomial is given by G=[1, g 0 /g 1 ], where g 0 =[1011] (feedback) and g 1 =[1101] (feedforward). The structure of the Turbo encoder used in LTE can be seen in Fig. 9. Figure 9: Structure of the LTE Turbo encoder (dotted lines are for trellis termination) As can be seen in Fig. 9, the output of the LTE Turbo encoder consists of three parts, a systematic bit and two parity bits. The systematic bit (X k ) is the untouched input bit. The first parity bit (Z k ) is the output of the first convolutional encoder with the original input (C k ) as input and the second parity bit (Z k) is the output of the second convolutional encoder after

10 10 Shahram Zarei interleaving (by the Turbo code internal interleaver) of the input bit (C k) as its input. For trellis termination the tail-bits X k are inserted. Turbo coding is used in LTE for following channels: UL-SCH (Uplink Shared CHannel), DL- SCH (Downlink Shared CHannel), PCH (Paging CHannel) and MCH (Multicast CHannel) Turbo encoders internal contention-free interleaver The internal contention-free interleaver is one of the key parts of the LTE Turbo encoder. The main difference between the Turbo encoders in LTE and UMTS is that the interleaver in LTE is in contrast to the one in UMTS contention-free. Contentions occur, when parallel working processeses try to write or read to/from the same memory address simultaneously. Because the two SISO (Soft-Input Soft-Output) MAP decoder engines of the Turbo decoder use such processes, the contention-free concept becomes survival for designing the Turbo encoders internal interleaver efficiently. A contention-free interleaver π(i), 0 i K should satisfy the following inequality, where W is the window size and 0 νw, u 1 0, u 2 < M for all U 1 U 2 : ψ(u1w + ν) ψ(u2w + ν) = (8) W W The above inequality satisfies for both interleaver ψ = π and deinterleaver ψ = π 1 and indicates, that the memory addresses (both sides of the inequality), accessed by M processes on the ν th step must be different. There were two candidates for the LTE internal interleaver: Almost Regular Permutation (ARP) and Quadrature Permutation Polynomial (QPP), which are very similar. However QPP is chosen for LTE because it offers more parallelism. The QPP interleaver for a block size of K is defined as following: π(i) = (f 1 i + f 2 i 2 ) mod K (9) Where i is the input index and π(i) the output index and f 1 and f 2 are permutation parameters, which can be get from the standard Turbo decoder in LTE The Turbo decoder in LTE is based on two SISO (Soft-Input Soft-Output) decoders, which work together in an iterative manner. Each SISO decoder has two inputs, namely a normal input and an a-priori Log Likelihood Ratio (LLR) input, and two outputs, an a-posteriori LLR and an extrinsic LLR. In each iteration the (de)interleaved version of the extrinsic output of a decoder is used as the a-priori information for the other decoder. Typically after 4 to 8 iterations, the a-posteriori LLR output can be used to obtain the final hard decision estimates of the information bits. The structure of the LTE Turbo decoder can be seen in Fig. 10.

11 Channel Coding and Link adaptation 11 Figure 10: Turbo decoder 3.3 Rate matching in LTE As can be seen in Fig. 9, the mother code rate of LTE Turbo encoder is 1/3. In order to get other code rates, if desired, repetition or puncturing has to be performed, which both are done by a rate matching module. The rate matching module consists of three so-called sub-block interleavers for the three output streams of the Turbo encoder core and a bit selection and pruning part, which is realized by a circular buffer. The sub-block interleaver is based on the classic row-column interleaver with 32 columns and a length-32 intra-column permutation. The bits of each of the three streams are written row-by-row into a matrix with 32 columns (number of rows depends on the stream size). Dummy bits are padded to the front of each stream to completely fill the matrix. After a column permutation, bits are read out from the matrix column-by-column. The column permutation of the sub-block interleaver is given as following: [0, 16, 8, 24, 4, 20, 12, 28, 2, 18, 10, 26, 6, 22, 14, 30, 1, 17, 9, 25, 5, 21, 13, 29, 3, 19, 11, 27, 7, 23, 15, 31] For example if column 1 has to be read orginally, after permutation column 16 will be read. Figure 11: Turbo encoder with rate matching module

12 12 Shahram Zarei Circular buffer in LTE The circular buffer is the most important part of the rate matching module, making puncturing and repetition of the mother code possible. The structure of the Turbo encoder with rate matching module is shown in Fig. 11 The interleaved systematic bits are written into the circular buffer in sequence, with the first bit of the interleaved systematic bit stream at the beginning of the buffer. The interleaved and interlaced parity bit streams are written into the buffer in sequence, with the first bit of the stream next to the last bit of the interleaved systematic bit stream. Fig. 12 shows the internal structure of the circular buffer. The white cells contain systematic bits and red and blue cells contain parity 0 and parity 1 bits, respectively. Green cells are the RV (Redundancy Version) points, which are considered with more details in HARQ section. The number of coded bits (depending on the code rate), are read out serially from a certain starting point specified by RV points in the buffer. If the end of the buffer is reached and more coded bits are needed for the transmission (in the case of a code rate smaller than 1/3), the transmitter wraps around and continues at the beginning of the buffer. Figure 12: Structure of a circular buffer for the LTE rate macthing module 3.4 HARQ and redundancy versions HARQ, which stands for Hybrid ARQ, is an error correction mechanism in LTE based on retransmission of packets, which are detected with error. The functionality of the HARQ can be seen in Fig. 13. The transmitted packet arrives after a certain propagation delay in receiver. Receiver produces either an ACK for the case of error-free transmission or a NACK, if some errors are detected. The ACK/NACK is produced after some processing time and sent back to transmitter and arrives there after a propagation delay. In the case of a NACK, after a certain processing delay in transmitter, the desired packet will be sent again. The bits, which are read out from the circular buffer and sent in each retransmission are different and depend on the position of the RV (Redundancy Version). There are four RVs (0, 1, 2, 3), which define the position of the starting point, where the bits are read out from the circular buffer. The RVs can be seen in the Fig. 12 and are presented as green cells. As can be seen in Fig. 12, in the first retransmission, more systematic bits are sent and with the progressing number of retransmissions, RV becomes higher and therefore less systematic and more parity bits are read out from the circular buffer for the retransmission.

13 Channel Coding and Link adaptation 13 Figure 13: HARQ mechanism in LTE 3.5 Convolutional coding for control channels in LTE In control channels like Physical Downlink Control CHannel (PDCCH) and Physical Broadcast CHannel (PBCH) a convolutional code is used. The reason is that the code blocks are much more smaller than in data channels and for small blocks the Turbo codes internal interleaver does not work efficiently. Because of the small number of bits carried in PDCCH and PBCH and in order to avoid overhead, the tail-biting approach instead of inserting tail bits is used as terminating method. The initial value of the shift register of the encoder shall be set to the values corresponding to the last 6 information bits in the input stream so that the initial and final states of the shift register are the same. The decoder can be either a circular Viterbi algorithm or a MAP algorithm. The rate-matching for the convolutional code in LTE uses a similar circular buffer method as for the Turbo code. 4 Summary LTE uses Adaptive Modulation and Coding (AMC) as the link adaptation technique to adapt transmission parameters, modulation scheme and code rate dynamically to the channel. The higher the channel quality, the higher is the used modulation order and code rate. Channel quality in downlink is measured in UE using the reference symbols. Upon this measurement, so-called CQIs, which are channel quality indicators are generated and sent to enodeb. Each CQI value corresponds to a specific modulation scheme and a specific code rate, which are selected by enodeb for the downlink transmission. LTE uses both convolutional and Turbo encoding for control and data channels respectively. With Turbo codes performance up to a few tenth of a db from the Shannons limit can be achieved. One of the highlights of the Turbo encoder in LTE is the internal contention-free interleaver, which makes it possible, that the Turbo decoder works with very high bit rates. Another feature of the LTE Turbo encoder is the concept of the circular buffer in the rate matching module. It allows a simple realization of rate matching in respect of HARQ.

14 14 Shahram Zarei References [1] S. Sesia, I. Toufik, M. Baker: LTE - The UMTS Long Term Evolution: From Theory to Practice, John Wiley and Sons, [2] W. Koch: Lecture script "Fundamentals of Mobile Communications", University of Erlangen-Nuremberg, [3] J. Cheng, A. Nimbalker, Y. Blankenship, B. Classon, K. Blankenship: Analysis of Circular Buffer Rate Matching for LTE Turbo code [4] M. Stambaugh: HARQ Process Boosts LTE communications, Agilent technologies, [5] C. Berrou, A. Glavieux, P. Thitimajshima: Near Shannon limit error-correcting coding and decoding: Turbo codes (1), [6] Third Generation Partnership Project; Technical Specification Group Radio Access Network; Evolved Universal Terrestrial Radio Access (E-UTRA); Multiplexing and channel coding (Release 8)