3620

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 9, NO. 12, DECEMBER 2010

Differential Quasi-Orthogonal Space-Frequency Trellis Codes

Jorge Flores, Jaime Snchez, Member, IEEE, and Hamid Jafarkhani, Fellow, IEEE

AbstractTwo rate-one differential quasi-orthogonal spacefrequency trellis codes (DQOSFTCs) for multiple-input multipleoutput (MIMO) orthogonal frequency division multiplexing (OFDM) channels are proposed. The DQOSFTCs are systematically constructed within an OFDM symbol period, by combining unitary quasi-orthogonal trellis codes and differential modulation over the frequency domain (DF-QOSFTC) or time domain (DT-QOSFTCs). Besides multipath diversity, our DQOSFTCs achieve high-coding gain and simple decoding when the channel state information is not available at both the transmitter and the receiver. Simulation results show that our proposed codes signicantly outperform the existing differential space-frequency trellis codes. Index TermsMIMO-OFDM, differential modulation, quasiorthogonal codes, space-frequency codes, trellis codes.

I. I NTRODUCTION PACE-FREQUENCY coded multiple-input multipleoutput orthogonal frequency division multiplexing (MIMO-OFDM) systems are capable of achieving maximum diversity over a frequency selective channel (FSC) [1]. However, acquiring knowledge of the channel state information (CSI) for an FSC with many taps is prohibitively complex. As a result, exploiting the available spatial and frequency diversities, if neither the transmitter nor the receiver knows the CSI, have been recently addressed in the literature. In order to achieve the space-time (ST) and space-frequency (SF) diversity, differential and noncoherent schemes were proposed in [2][12]. In many of these schemes, maximum diversity and good performance can be achieved by dividing an OFDM symbol into several subcarriers groups, and perform differential encoding and decoding between adjacent groups. However, the existing differential techniques have several drawbacks. For example, they have a large code size, which exponentially increases the coding and decoding complexity. Moreover, by trying to increase the data rate, they have a low coding gain. In addition, these implementations require constant channel frequency response (CFR) from group to group, which lead to severe error oor in FSCs. Differential SF trellis codes (DSFTCs) presented in [10], can obtain rate-one (1 symbol per subcarrier), spatial diversity, and a simple decoding comManuscript received March 23, 2010; revised July 20, 2010; accepted September 11, 2010. The associate editor coordinating the review of this letter and approving it for publication was X. Wang. J. Flores and J. Snchez are with the Department of Electronic and Telecommunications, CICESE Research Center, 3918 Carr. TijuanaEnsenada Zona Playitas , Ensenada, B. C. 22860 Mxico (e-mail: {jtroncos, jasan}@cicese.mx). H. Jafarkhani is with the Center for Pervasive Communications & Computing, University of California at Irvine, Irvine, CA 92697 USA (e-mail: hamidj@uci.edu). Digital Object Identier 10.1109/TWC.2010.102210.100463

plexity. Even though the DSFTCs increase the coding gain, the error oor is still signicant. The differential codes proposed in this letter, called differential quasi-orthogonal SF trellis codes (DQOSFTCs), solve these drawbacks. The proposed DQOSFTCs achieve high coding gain, and are capable of exploiting both spatial and frequency diversity over a MIMO-OFDM system with lack of CSI at the transmitter and at the receiver. In order to guarantee both rate-one and full-diversity, we use the structure of the generalized quasi-orthogonal space-time block code (QOSTBC) derived in [13]. We stress, however, that the construction in [13] does not take into account the coding gain, and it is not suitable for differential encoded process. In summary, the contributions of this letter are the following: 1) In order to get rate-one and high-coding gain, we obtain a sufcient number of full-diversity unitary quasi-orthogonal codes, and following a similar procedure as in [14], we perform set partitioning. Then, we systematically design the unitary quasi-orthogonal SF trellis codes (QOSFTCs). 2) We propose to perform the differential encoding in 2 ways: a) over the frequency domain, b) over the time domain. 3) According to the number of orthogonal matrices (subblocks) within a QOSFTC, we divide the OFDM subcarriers into equidistant groups. In addition, we take advantage of orthogonality of the inner subblocks of the QOSFTCs, by using a simple Maximum-Likelihood (ML) decoder without requiring CSI at the receiver, which is formed by a differential decoder and a Viterbi decoder. Therefore, we obtain a simple differential encoding process, and the assumption of a constant CFR from group to group can be relaxed. Besides being computationally efcient and easy to implement, we show through numerical simulations that in the presence of FSCs, our proposed codes signicantly outperform the existing DSFTCs. II. S YSTEM M ODEL Consider a system implemented with transmit and receive antennas. We assume no spatial fading correlation exists between antennas. Each transmit antenna employs an subcarrier OFDM modulator. We denote the transmitted space = frequency code at the th OFDM symbol period by C ] [ , where c = [ (1) (2) . . . ( )] c1 . . . c C is transmitted from the th antenna ( = 1 . . . ), and () is the complex data transmitted at the th subcarrier ( = 1, . . . , ); superscript () denotes the vector transpose and C represents the complex eld of dimension . satises the power constraint C 2 = , Moreover, C 2 where is the Frobenius norm. In order to avoid the Inter Symbol Interference (ISI) which is produced by the multipath delay of the channel, a cyclic prex with the proper length is added to each OFDM symbol.

c 2010 IEEE 1536-1276/10$25.00

FLORES et al.: DIFFERENTIAL QUASI-ORTHOGONAL SPACE-FREQUENCY TRELLIS CODES

3621

We assume that the channel impulse response (CIR) between the transmit antenna and receive antenna has independent delay paths on each OFDM symbol and an arbitrary power delay prole, which is denoted as , () = 1 =0 , ( ) ( ), where represents the th path delay, and each fading coefcient , () at delay is a complex 2 Gaussian random variable with zero mean and variance . 1 2 For normalization purposes =0 = 1. The CFR at the th subcarrier between transmit antenna and receive antenna is expressed as ( ) 1 , () = , () exp 2 , (1)
=0

identity matrix and superscript () denotes the conjugate transpose. Let us denote by O the set of all unitary codewords . For the sake of providing the required cardinality = 2 , we introduce unitary rotation matrices U = diag(,1 , ,2 , . . . , ,2 ), = 0, 1, . . . , and diag() denotes a diagonal matrix, such that = {OU0 , . . . , OU }. Thus, we express an information codeword from the trellis encoder U . Since U is a unitary matrix, then = output as is also a unitary matrix. Due to the diagonal structure of (3), we can write the inner orthogonal matrices as 1 B (21 , 2 ) = A(21 , 2 ) diag(,21 , ,2 ), 2 (4) where = 1, . . . , . Then, we dene the th unitary codeword sent at period from a trellis encoder as 0 B, (1 , 2 ) . . . . . .. = . . , (5) . , . . 0 . . . B, (21 , 2 ) C22 , = 0, 1, . . . , 1, and = . where , 2 For convenience, we assume that = 2 for some integer . Note that the essential structure of the QOSTBC derived in [13] is preserved by the unitary matrix dened in (5), therefore the full-diversity of is still maintained. B. Differential modulation For simplicity, in the sequel, we consider a scenario with = 2 transmit antennas. In order to perform differential encoding keeping multipath diversity, we exploit the independence of the orthogonal matrices in (5) by partitioning the total subcarriers into groups of subcarriers. Then, we propose encoding differentially over the frequency domain (DF), and over the time domain (DT) as follows. 1) DF: We consider that the channel remains constant during one OFDM symbol period and changes independently from symbol to symbol. We also consider that the correlation between adjacent subcarriers is high, which is a reasonable assumption if is large enough. In the differential encoder, the 22 recursive construction of the unitary codeword , C is given by { = 0 I2 , C (6) , (S ) = B, (S )C,1 (S ), 1, where for the sake of brevity, we use the symbol S to denote the pairwise (21 , 2 ), = 1, 2, . . . , . 2) DT: In this case we assume that the channel is quasistatic over one OFDM symbol period and slowly varies C22 be the between adjacent OFDM symbols. Let , differential encoded matrix to be transmitted at the th OFDM symbol period. At the beginning of the transmission ( = 0), the differential encoder sends C0 , (21 , 2 ) = I2 for all = 0, 1 . . . , 1, where = 1, 2, . . . , . The th transmit codeword is generated as { I2 , = 0 C, (S ) = (7) 1 B, (S )C, (S ), 1.

where = / is the th path delay normalized and is the sampling rate of the OFDM system. Assuming perfect sampling time and carrier synchronization at the receiver, after matched ltering, removing the cyclic prex and FFT processing, the signal at th receive antenna and th subcarrier is given by
() = =1 where () is a circularly symmetric zero-mean and unitvariance Gaussian noise term at the th symbol duration. (), () + (),

(2)

III. D IFFERENTIAL SF T RELLIS C ODED M ODULATION In this section, we derive the rate-one unitary QOSTBC construction, and describe two differential encoding designs. We also show the systematic construction for the rate-one . DQOSFTCs and the transmitted codeword C A. Unitary quasi-orthogonal block codes The rst stage of information processing in the transmitter is the trellis encoder. Similar to the method in [14], depending on the input information bits and the current state in the trellis encoder, a codeword is selected from a constellation of possible unitary-QOSTBCs. Then, there is a transition towards the next state in the trellis, and the encoding process is performed again to obtain a frame with codewords. Let us consider the general class of QOSTBC dened in [13] as A(1 , 2 ) . . . 0 . . .. = 1 . . , (3) . . . 2 0 . . . A(21 , 2 ) C22 , and A( , ) where space goes horizontally, denotes the Alamouti code [15] for any indeterminate symbols , . In order to support a data rate of , dened as the number of bits per subcarrier use (b/s/Hz), it is necessary for a sequence of information bits to pick a codeword as (3) at each state of the trellis encoder. Then, we denote as the set of all possible information codewords, which must have a cardinality of = 2 . Furthermore, in the interest of performing differential encoding after trellis encoding, the information codewords to be transmitted must be = I2 , where I is the unitary matrices, such that

3622

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 9, NO. 12, DECEMBER 2010

Afterwards, the differentially encoded matrices are mapped to the same subblock of two adjacent time domain OFDM symbols. C. Subcarrier assignment After picking the unitary codeword (5) from the trellis encoder output and the differential encoding process, we map the codeword , to the OFDM-subcarrier groups. Then, we express a SF construction for two transmit antennas and subcarrier groups as [ = C C ,0 (1 , 2 ) . . . C, 1 (1 , 2 ) ,
C ,0 (3 , 4 ) . . . C, 1 (3 , 4 ) ,

Fig. 1.

Final set partitioning of sets (a) with 0 , and (b) with 1 .

C ,0 (21 , 2 ) . . . C, 1 (21 , 2 )

2 . where C D. Differential QOSFTCs

. (8)
Fig. 2. Set partitioning for sets O and P using QPSK.

In what follows, we propose rate-one differential QOSFTC designs for two transmit antennas, assuming a two-path fading channel model, i.e. = 2. However, it is noteworthy that the method described here can be extended in a straightforward way to > 2. 3 ), 3 = (1 3 ), 1) Codeword sets: Let 1 = (1 + 2 = (2 + 4 ) and 4 = (2 4 ) be the symbols transmitted in (5) by setting = 2, where 1 , 2 belong 4 belong to the rotated to a M-PSK constellation , and 3 , constellation . Also, let = / be the optimal rotation that provides the maximum coding gain for the code in (5) [13]. In the sequel, we asume that the system employs a QPSK constellation. Therefore, = 2 denotes a QPSK symbol, where = {0, 1, 2, 3} represents the index of the corresponding QPSK symbol. Thus, we obtain only 128 unitary codewords in O. Therefore, in order to get a data rate of = 2 b/s/Hz, we need two unitary matrices U0 , U1 such that = {OU0 , OU1 } and = 256. Due to the symmetry in (5), if (1 , 2 ) and ( 3 , 4 ) , we get another set of 128 different unitary codewords called P. Let 1 and 2 be the rotation angles 3 , 4 ), respectively. Then, for the symbols (1 , 2 ) and ( we assign 0 = {1 = 0, 2 = /4} for the 128 unitary codewords belonging to set O, and 1 = {1 = /4, 2 = 0} for the 128 unitary codewords that belong to set P. Thus, we obtain a new set = {PU0 , PU1 } such that = 256. 2) Design criteria: Here we borrow the design criteria for differential space-time-frequency codes of [7], which are identical to those proposed for differential super-orthogonal space-time trellis codes over at channels in [11]. Let 1 and 2 be two codewords as dened in (5) so that: i) Full diversity is achieved if the difference matrix D = 1 2 has full rank over all possible pairs of distinct codewords 1 and 2 . ( ) ii) The minimum of the determinant det D D 1 = 2 corresponds to the coding gain distance (CGD) and must be maximized.

3) Set partitioning: Note that the codeword structure in (5) guarantees the full diversity criterion. Moreover, the minimum CGD between codewords at each level of an optimal set partitioning must be maximized. In order to avoid expanding the original QPSK constellation, we need to search for the optimal values of U ( = 0, 1). In addition, the design criteria for the codewords in set or must be satised. With numerical search, we obtain the optimal matrices U0 = diag(1, , 1, 1), U1 = diag(1, , 1, 1). As a consequence, each set and contains 256 rate-one fulldiversity unitary quasi-orthogonal codes using QPSK. In order to design a trellis code, we partition the sets following a procedure similar to the procedure in [14]. Due to the lack of space, we only show the nal set partitioning. Fig. 1 shows the set partitioning for sets and with QPSK, while Fig. 2 shows a partial set partitioning for O and P. 4) Differential trellis encoding: We now show how to use the proposed set partitioning schemes to systematically design rate-one full-diversity differential QOSFTCs. A key concept is that those codewords that do not belong to the same set are assigned to different states. Moreover, we must assign codewords diverging (or merging) into a state such that both difference matrices have full rank, and the CGD between all pairs of codewords must be the largest CGD. In Fig. 3 we propose 4-state and 8-state trellises containing branches with 64 and 32 parallel transitions, respectively. According to the current state and the 8 input bits, the trellis {, } based on (5). encoder builds the th codeword , Full-diversity and high-coding gain QOSFTCs with different number of states can be systematically designed using both the corresponding set partitioning and the design rules described above. , in order to build a After constructing the codeword , DF-QOSFTC or a DT-QOSFTC, we rst perform the differential encoding using (6) or (7), respectively. Then, at the th OFDM symbol period SF code as in (8) for [ we form a = C ( , ) . . . C = 2, such that C 1 2 ,0 , 1 (1 , 2 ) ,

FLORES et al.: DIFFERENTIAL QUASI-ORTHOGONAL SPACE-FREQUENCY TRELLIS CODES

0

3623

10

OFDM Symbol Error Rate (SER)

10

10

2ray uniform
10
3

TU fDn=0.0025 8state Coherent 8state DSFTC [10] 4state DFQOSFTC 8state DFQOSFTC 4state DTQOSFTC 8state DTQOSFTC
12 14 16 18

10

Fig. 3.

QOSTFTCs using QPSK (a) 4-state, (b) 8-state.

10
5

C ,0 (3 , 4 )

...

] C . , 1 (3 , 4 )

10

SNR [dB]

20

22

24

26

28

30

E. ML Decoder The successful differential decoding of the proposed DFQOSFTCs and DT-QOSFTCs depends on the assumption that the fading channel remains constant over at least four adjacent subcarriers or two successive OFDM symbols, respectively. It is worth mentioning that both, orthogonality and independence of the inner building matrices in the proposed codes, make it possible to simplify the complexity of the decoding process by separating the decoding of these subblocks. In order to estimate the information subblock B , (21 , 2 ), no CSI is required at the receiver. The proposed differential codes are decoded using Maximum-Likelihood (ML) decoding, comprised by the differential decoder derived in [3] and a Viterbi decoder. Note that to perform differential decoding for the DT-QOSFTCs, two consecutive received OFDM symbols are required. IV. S IMULATION R ESULTS In this section, we provide simulation results for the performance of the differential QOSFTCs proposed. We assume a 128-tone OFDM system with two transmit antennas, a single receive antenna, and a total bandwidth of 1 MHz. We add a cyclic prex of 20 s to avoid the ISI. The performance curves are described by means of OFDM symbol error rate (SER) versus the receive SNR. We simulate the system over the following channel scenarios: 1) A quasi-static channel with 2-path uniform power delay prole, which changes independently for each OFDM symbol. The delay between these 2 paths is one OFDM sample duration. 2) The channel is quasi-static over one OFDM symbol period and slowly changes between adjacent OFDM symbols by varying the normalized Doppler frequencies , with a) = 0.0025 and b) = 0.0125, which correspond to mobile speeds of 6 and 30 m/s, respectively. To properly model an FSC, we adopt a typical urban (TU) six-path power delay prole. In order to validate the work we have carried out, we then compare the performance of our proposed DF-QOSFTCs and DT-QOSFTCs, to that of the 8-state DSFTC in [10], which is based on group codes with the same rate =2 b/s/Hz. We have added a comparison with coherent QOSFTCs and the result is a 3 dB difference, as happens with most well-designed

Fig. 4. SER performance for =2 b/s/Hz under the 2-ray and TU power delay proles.
10
0

OFDM Symbol Error Rate (SER)

10

10

10

TU fDn=0.0025 TU f =0.0125
Dn

10

8state Coherent 8state DSFTC [10] 8state DFQOSFTC 8state DTQOSFTC

12 14 16 18

10

SNR [dB]

20

22

24

26

28

30

Fig. 5. SER performance for =2 b/s/Hz under the TU power delay prole.

differential schemes. As can be seen from Figs. 4 and 5, our proposed codes all considerably outperform the existing DSFTC. Fig. 4 provides SER performance for the channel scenarios 1) (showed in dashed lines), and 2a) (showed in solid lines). Fig. 4 shows that the DF-QOSFTCs and DT-QOSFTCs over channel scenarios 1) and 2a), respectively, achieve the full-diversity order of 4. Note that we have designed codes for = 2 and therefore the maximum achievable diversity is four, independent of the actual number of taps in the system. However, although DF-QOSFTCs also reach coding gain under the channel scenario 2a), at high SNRs, they suffer a diversity loss. This is because the success of the DFQOSFTCs depends on a high correlation between adjacent subcarriers, which is not preserved in hostile FSCs. This effect can be mitigated either by increasing the number of subcarriers or implementing codes for > 2. Thus, the number of subcarrier groups is increased leading to a better multipathdiversity gain. Next, we compare the performance of the 8-state proposed codes versus existing 8-state DSFTC by varying . We can see in Fig. 5 that both DF-QOSFTC and DSFTC have a robust

3624

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 9, NO. 12, DECEMBER 2010

performance in the case of time-varying channels. However, time variations of the channel affect the diversity of the DTQOSFTC, because of the differences between the channel gain of subcarrier groups for adjacent OFDM symbols. Nevertheless, the DT-QOSFTC notably outperforms all other codes. Moreover, the transmission efciency for the DT-QOSFTCs is greater than that achieved in the frequency domain by for example DF-QOSFTCs and DSFTC. In addition, under practical fading channels, signicant gains can be achieved by increasing the number of states in a DT-QOSFTC and the simple decoding complexity is hold. V. C ONCLUSIONS We have proposed rate-one differential space-frequency trellis codes for MIMO-OFDM systems, where CSI is not available at the transmitter and at the receiver. The proposed differential codes have been designed based on a generalized class of unitary quasi-orthogonal space-time block codes. Also by trellis encoding and grouping of OFDM-subcarriers, our proposed codes achieve full-diversity and high-coding gain over practical FSCs. Simulation results show that, compared to the existing DSFTC, the proposed codes provide better performance in simulated channel scenarios. Moreover, the independence of inner orthogonal matrices in the proposed codes, allows a reduced encoding and decoding complexity. Furthermore, because the differentially encoded codewords are transmitted within one OFDM symbol period, we obtain a reduced decoding delay. The provided examples have only been able to achieve a maximum diversity order of four with two transmit antennas, but it is straightforward to design similar codes for more than two taps, and two transmit antennas. This, of course, is at the expense of increased decoding complexity. In addition, unlike differential and noncoherent SF schemes in the literature, our proposed codes achieve rate one and can be designed in a simple systematic way.

R EFERENCES
[1] W. Su, Z. Safar, and K. Liu, Towards maximum achievable diversity in space, time, and frequency: performance analysis and code design," IEEE Trans. Wireless Commun., vol. 4, no. 4, pp. 1847-1857, July 2005. [2] V. Tarokh and H. Jafarkhani, A differential detection scheme for transmit diveristy," IEEE J. Sel. Areas Commun., vol. 18, no. 7, pp. 1169-1174, July 2000. [3] B. L. Hughes, Differential space-time modulation," IEEE Trans. Inf. Theory, vol. 46, no. 7, pp. 2567-2578, Nov. 2000. [4] B. Hochwald and W. Sweldens, Differential unitary space-time modulation," IEEE Trans. Wireless Commun., vol. 48, no. 12, pp. 2041-2052, Dec. 2000. [5] M. Tao, High rate trellis coded differential unitary space-time modulation via super unitarity," IEEE Trans. Wireless Commun., vol. 5, no. 12, pp. 3350-3354, Dec. 2006. [6] M. Borgmann and H. Blcskei, Noncoherent space-frequency coded MIMO-OFDM," IEEE J. Sel. Areas Commun., vol. 23, no. 9, pp. 17991810, Sep. 2005. [7] Q. Ma, C. Tepedelenlioglu, and Z. Liu, Differential space-timefrequency coded OFDM with maximum multipath diversity," IEEE Trans. Wireless Commun., vol. 4, no. 5, pp. 2232-2243, Sep. 2005. [8] Y. Zhu and H. Jafarkhani, Differential modulation based on quasiorthogonal codes," IEEE Trans. Wireless Commun., vol. 4, no. 6, pp. 3018-3030, Nov. 2005. [9] T. Himsoon, W. Su, and K. J. R. Liu, Single-block differential transmit scheme for broadband wireless MIMO-OFDM systems," IEEE Trans. Signal Process., vol. 54, no. 9, pp. 3305-3314, Sep. 2006. [10] Z. Hong, L. Thibault, and L. Zhang, Differential space-frequency trellis codes," IEEE Trans. Wireless Commun., vol. 5, no. 10, pp. 2664-2668, Oct. 2006. [11] Y. Zhu and H. Jafarkhani, Differential super-orthogonal space-time trellis codes," IEEE Trans. Wireless Commun., vol. 5, no. 12, pp. 36343643, Dec. 2006. [12] Q. Li, K. H. Li, and K. C. Teh, Noncoherent space-frequency codes for broadband MIMO systems over frequency-selective fading channels," in Proc. IEEE VTC Spring, 2008, pp. 554-558. [13] F. Fazel and H. Jafarkhani, Quasi-orthogonal space-frequency and space-time-frequency block codes for MIMO OFDM channels," IEEE Trans. Wireless Commun., vol. 7, no. 1, pp. 184-192, Jan. 2008. [14] H. Jafarkhani and N. Seshadri, Super-orthogonal space-time trellis codes," IEEE Trans. Inf. Theory, vol. 49, no. 4, pp. 937-950, Apr. 2003. [15] S. Alamouti, A simple transmit diversity scheme for wireless communications," IEEE J. Sel. Areas Commun., vol. 16, no. 8, pp. 1451-1458, Oct. 1998.