1890

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 52, NO. 11, NOVEMBER 2004

Performance of Coded OQPSK and MIL-STD SOQPSK With Iterative Decoding

Lifang Li and Marvin K. Simon, Fellow, IEEE
AbstractWe show that military standard (MIL-STD) shapedoffset quadrature phase-shift keying (SOQPSK), a highly bandwidth-efcient constant-envelope modulation, can be represented in the form of a cross-correlated trellis-coded quadrature modulation. Similarly, we show that offset QPSK (OQPSK) can be decomposed into a degraded trellis encoder and a memoryless mapper. Based on the representations of OQPSK and MIL-STD SOQPSK as trellis-coded modulations (TCMs), we investigate the potential coding gains achievable from the application of simple outer codes to form a concatenated coding structure with iterative decoding. For MIL-STD SOQPSK, we describe the optimum receiver corresponding to its TCM form and then propose a simplied receiver. The bit-error rate (BER) performances of both receivers for uncoded and coded MIL-STD SOQPSK are simulated and compared with that of OQPSK and Feher-patented QPSK (FQPSK). The asymptotic BER performance of MIL-STD SOQPSK is also analyzed and compared with that of OQPSK and FQPSK. Simulation results show that, compared with their uncoded systems, there are signicant coding gains for both OQPSK and MIL-STD SOQPSK, obtained by applying iterative decoding to either the parallel concatenated coding scheme or the serial one, even when very simple outer codes are used. Index TermsBit-error rate (BER), iterative decoding, offset quadrature phase-shift keying (OQPSK), trellis-coded modulation (TCM).

I. INTRODUCTION FFSET quadrature phase-shift keying (OQPSK) is a constant-envelope modulation that has no 180 phase shifts, and therefore has a much higher spectral containment than nonoffset QPSK when transmitted over bandlimited nonlinear channels. To further bandlimit an OQPSK signal, shaped OQPSK (SOQPSK) was introduced by Dapper and Hill [1] in the early 1980s, and its initial version was referred to as MIL-STD SOQPSK after it was adopted as part of a military standard. The frequency-shaping pulse for MIL-STD SOQPSK in its continuous phase modulation (CPM) representation is rectangular, and it lasts one bit interval [2]. Later on, more spectrally efcient versions of SOQPSK were developed by Hill [3], [4], and these variants are comparable to or even better than Feher-patented QPSK (FQPSK) [5] with regard

Paper approved by W. E. Ryan, the Editor for Modulation, Coding, and Equalization of the IEEE Communications Society. Manuscript received October 29, 2003; revised May 12, 2004. This work was performed at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration. The authors are with the Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109-8099 USA (e-mail: marvin.k.simon@ jpl.nasa.gov). Digital Object Identier 10.1109/TCOMM.2004.836557

to spectral and/or power efciency [3], [4], [6], [7].1 These variants replace the linear variation of the phase over the single bit interval inherent to MIL-STD SOQPSK and analogous to that of minimum-shift keying (MSK), with other variations that extend over several bit intervals, and as such, are analogous to partial-response types of CPM systems such as Gaussian MSK (GMSK). Our focus in this paper is on MIL-STD SOQPSK, and thus, the more recent work of Hill and others will not be further elaborated on.2 Cross-correlated trellis-coded quadrature modulation (XTCQM) was introduced by Simon and Yan [8] as a generic modulation scheme, containing both memory and cross-correlation between the inphase (I) and quadrature (Q) channels. One specic embodiment of XTCQM is FQPSK [9][13]. In this paper, we show how MIL-STD SOQPSK can also be represented in the form of an XTCQM. Analogous to FQPSK, the representation of MIL-STD SOQPSK in the form of XTCQM allows identication of an optimum receiver for it, and allows for its inherent memory to be used in the iterative decoding of its coded and interleaved systems.3 We describe such an optimum receiver and investigate the potential improvement in power efciency obtained from exploring the inherent memory of MIL-STD SOQPSK in a coded system with iterative decoding. Furthermore, we introduce a similar representation for OQPSK to that of XTCQM for MIL-STD SOQPSK. Based on this representation of OQPSK and the XTCQM representation of MIL-STD SOQPSK, we present a symbol-by-symbol mapping for both OQPSK and MIL-STD SOQPSK that is performed directly on the input I and Q data sequences in every symbol (2-bit) interval. This direct symbol-by-symbol mapping results in a clear interpretation for MIL-STD SOQPSK as being composed of a cross-correlated trellis encoder and a memoryless mapper and, for OQPSK, a degraded trellis encoder and a memoryless mapper. Such decomposition of OQPSK makes it possible to apply iterative decoding to coded OQPSK, where the degraded trellis code of OQPSK, after being remapped to its recursive version, can be viewed as an inner code of a concatenated coding structure. The performance of coded OQPSK with iterative decoding provides a lower bound to that of coded MIL-STD SOQPSK and FQPSK. In an effort to reduce receiver complexity while maintaining reasonable performance, we also propose a simplied receiver for MIL-STD SOQPSK. This simplied receiver only requires
1Note

that MIL-STD SOQPSK and its variants are unlicensed technologies.

2Sufce it to say that the ideas presented in this paper can be extended to those

more bandwidth-efcient SOQPSK variants. 3Note that iterative decoding of coded CPM waveforms in general can be found in [14][16] and the references therein.

0090-6778/04$20.00 2004 IEEE

LI AND SIMON: PERFORMANCE OF CODED OQPSK AND MIL-STD SOQPSK WITH ITERATIVE DECODING

1891

half of the matched lters used in the optimum receiver, and it has the same complexity as the simplied receiver for FQPSK [17]. We then simulate the bit-error rate (BER) performance of the optimum receiver and the simplied receiver for uncoded and coded MIL-STD SOQPSK, and compare it with that of OQPSK and FQPSK. In the coded case, we investigate the serially concatenated system with two very simple codes of different rates as outer codes and the parallel concatenated (turbo-like) system without any outer codes. Simulation results show that, compared with the uncoded case, there are significant coding gains for both parallel and serially concatenated systems, even with these simple codes. The remainder of the paper is organized as follows. In Section II, we show the decompositions of OQPSK and MIL-STD SOQPSK by describing a time-invariant symbol-interval trellis representation for them obtained from their CPM forms, and give equivalent transmitter implementations for them. Based on their symbol-interval trellis representations, in Section III, we present the explicit trellis-coded modulation (TCM) forms for OQPSK and MIL-STD SOQPSK. The optimum receiver and the simplied receiver for MIL-STD SOQPSK are described in Section IV. In Section V, we give the analytical asymptotic BER performance as well as the simulated BER results of uncoded MIL-STD SOQPSK with both receivers, and compare them with those of OQPSK and FQPSK. In Section VI, we describe the coded OQPSK and MIL-STD SOQPSK systems with iterative decoding and present the simulated BER performance of them, followed by our conclusions in Section VII. II. DECOMPOSITIONS OF OQPSK AND MIL-STD SOQPSK In order to show the time-invariant symbol-interval trellis representations of OQPSK and MIL-STD SOQPSK, as well as the corresponding equivalent transmitter implementations, we rst give a brief review of the CPM representations of OQPSK and MIL-STD SOQPSK, and describe an eight-state bit-interval trellis diagram of OQPSK based on its CPM representation. A. CPM Representations of OQPSK and MIL-STD SOQPSK It is known that a conventional OQPSK signal can be represented as a full-response CPM signal in the form of [18]

binary ( 1) data sequence by [18] (2) Detailed explanations on how (2) can be obtained through an eight-state (3-bit-state) trellis diagram of OQPSK (which will be described in the next subsection) are given in [19]. Note that the CPM representation of OQPSK can be implemented with the cascade of a precoder satisfying (2) and a conventional CPM modulator, such as in [18, Fig. 2-7]. To improve bandwidth efciency, SOQPSK introduces pulse shaping into the above CPM representation. In particular, MIL-STD SOQPSK uses a rectangular frequency pulse in each ]. Equivalently, the phase pulse bit interval [i.e., varies linearly with time over each bit interval, and as such, is still characterized by a full-response precoded CPM as that of OQPSK. B. Eight-State Bit-Interval Trellis Diagram of OQPSK We now describe an eight-state bit-interval trellis diagram of OQPSK based on its CPM representation, which can be used to obtain (2) as well as to provide a means for demodulation of OQPSK using a Viterbi algorithm (VA). The 3-bit trellis state variable for describing the CPM representation of OQPSK is dened as follows. The rst bit corresponds to whether the coming input bit, i.e., , corresponds to an even interval (I) or odd interval (Q). In particular, we choose a 1 if the incoming bit will be assigned to the I channel (even bit interval), and a 0 if the incoming bit will be assigned to the Q channel (odd bit interval). Note that successive input bits will be alternately assigned to even (I) and odd (Q) intervals. Therefore, states starting with a 0 can only transition to states starting with a 1, and vice versa. The second and third bits of a trellis state correspond to the current phase state, which is represented by the current I and Q bits, respectively. Specically, assuming a conventional Gray code mapping, the phase states , , , and are assigned the bit mappings (in the form of IQ) 00, 10, 11, 01, respectively. Note that at most one bit (I or Q) of the phase state can change during each state rad transition, i.e., the phase change is constrained to be or 0 in each bit interval. For example, the phase state 00 can only transition to phase states 10 and 01 , , depending on whether the or remain in phase state 00 incoming bit is an I bit or a Q bit, and whether its value is 1 rad, rad, or 0. The corresponding phase changes are and 0, respectively, which implies that the corresponding s are 1, 1, and 0, respectively. Fig. 1 is the eight-state trellis diagram illustrating the transitions from state to state in accordance with the above. The branches of the trellis are labeled with the value of that results in the transition due to the corresponding phase-state change. The input bits of the upper and lower branches emerging from each trellis state are 0 and 1, respectively. It can be easily veried that the value of on each branch is in accordance with (2). A detailed example is given in [19]. As shown in Fig. 1, in each bit (half-symbol) interval, the eight-state for an trellis diagram of OQPSK gives the corresponding

where and denote the energy and duration of a bit, respectively, is the carrier frequency, and is an arbitrary phase constant that, without loss of generality, can be set to zero. In is the phase-modulation process that can be addition, expressed as (1) where, for unshaped OQPSK, the modulation index , the normalized phase pulse is simply a step function in each bit period (equivalently, the frequency pulse is a delta function, i.e., ), and the th element of the effective ternary ( 1, 0, 1) data sequence , , is related to the true input

1892

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 52, NO. 11, NOVEMBER 2004

, denoted by , 0, 1, waveforms that can result for , 0, 1, and , 2, 3. In addition, the expressions for 0, 1, 2, 3, in each symbol interval are

Similarly, for MIL-STD SOQPSK, there are eight possible waveforms that can result for , denoted by , , and eight possible waveforms that can result , denoted by , . The expressions for , 0, 1, 2, 3, and , 0, 1, 2, 3, are for

Fig. 1. Eight-state bit-interval trellis diagram of OQPSK.

input bit associated with the transition between two states. Based on this trellis diagram, in Section II-C, we consider transitions between the four phase states corresponding to a pair of input bits in each symbol interval. C. Time-Invariant Symbol-Interval Trellis Representations of OQPSK and MIL-STD SOQPSK Without loss of generality, assume that in each symbol interval, the rst bit of the input pair is always an I bit. Then, given the eight-state trellis diagram of OQPSK in Fig. 1, we can , easily obtain the trellis between the four phase states , , and , which is illustrated in Fig. 2. Note that we have drawn the trellis in expanded form with each transition interval (now 2 bits in duration) showing the transitions leaving from one of the four phase states. In Fig. 2, each branch is now labeled with a pair of output values, i.e., , . The corresponding pair of input bits is the same as the pair of bits representing the terminating phase state. Given ) for each transition, there is a the pair of outputs ( , pair of waveforms ( , ) associated with it, which represents the pair of symbols synchronously transmitted on the I and Q chanis the initial phase of each transition indicated by nels. Here the starting phase state, and in each symbol interval

and, in addition, , , , 0, 1, 2, 3. and for OQPSK Graphical illustrations of each and MIL-STD SOQPSK are given in [19] and [20]. Note that, for each phase-state transition in Fig. 2, the corresponding , ) for MIL-STD SOQPSK is inwaveform pair ( dicated above the associated branch, and the waveform pair , ) for OQPSK is indicated below the branch. ( D. Equivalent Transmitter Implementations of OQPSK and MIL-STD SOQPSK

for OQPSK. For MIL-STD SOQPSK, we have

Given the 16 possible combinations of output pair and initial phase as shown in Fig. 2, it is easily veried that for OQPSK, there are only two possible waveforms that can , denoted by , 0, 1, and only four possible result for

Based upon the foregoing trellis representations of OQPSK and MIL-STD SOQPSK, as well as the labeling of their I and Q waveforms described in Section II-C, for both OQPSK and MIL-STD SOQPSK, we can express the indexes of the speand in each symbol cic waveforms transmitted for (2-bit) interval in terms of the two values in this interval and the phase state at the beginning of the interval (which itself depends on the previous values of ). Specically, corresponding and in the symbol interval to ( even) and phase state at the start of this interval, we have

LI AND SIMON: PERFORMANCE OF CODED OQPSK AND MIL-STD SOQPSK WITH ITERATIVE DECODING

1893

Fig. 2. Expanded (branch leaving each state) time-invariant symbol-interval phase trellis for OQPSK and MIL-STD SOQPSK.

and , where the binary-coded decimal (BCD) representations of and are if if if if for OQPSK, with denoting the XOR operation for binary (0, 1) data. For MIL-STD SOQPSK, we have if if

th symbol interval is simply and it will be in the th symbol interval, with the input come data and . Therefore, given the indexes of the cor, ], as indicated responding output waveform pair [ on each branch in the trellis representations of OQPSK and MIL-STD SOQPSK in Fig. 2, it is straightforward to express the indexes in terms of the input data. In particular, assume that , ] in the th symbol inthe output waveform pair is [ terval, and dene the indexes and by

Then, for MIL-STD SOQPSK, we have if if The BCD representations of and for OQPSK and MIL-STD SOQPSK can be easily veried from Fig. 2. Therefore, the OQPSK and MIL-STD SOQPSK transmitters can be equivalently implemented with the cascade of a precoder satisfying (2), a signal mapper for choosing I and Q waveforms based on the above considerations, and a quadrature modulator. Block diagrams of these equivalent transmitters are given in [19] and [20]. III. INTERPRETATION OF OQPSK AND MIL-STD SOQPSK AS TCM In Section II, we expressed for OQPSK and MIL-STD SOQPSK the indexes of the specic waveforms transmitted for and in terms of the two values in each symbol interval and the starting phase at the beginning of each interval. In this section, we will show for both OQPSK and MIL-STD SOQPSK that the indexes of the transmitted waveforms for and can be directly expressed in terms of the I and Q channel input binary (0, 1) data. Specically, in each symbol interval , we denote the I and Q input binary data as and , respectively. Note that the phase state in the

and for OQPSK

Graphical illustration of the implementations of MIL-STD SOQPSK and OQPSK based on the above mappings is given in Fig. 3. From Fig. 3(a), we see that MIL-STD SOQPSK can be clearly decomposed into a four-state trellis encoder and a memoryless signal mapper. This inherent four-state trellis encoder of MIL-STD SOQPSK has two binary (0, 1) inputs and and two waveform outputs , , where the trellis state is dened by the 2-bit sequence and . The trellis of this four-state encoder is exactly the one illus, ] indicated above each branch. trated in Fig. 2 with [ Since both the I and Q channel output waveform indexes depend on the cross-channel input data in addition to their own channel-input data, it is obvious that MIL-STD SOQPSK is a form of XTCQM. Similarly, from Fig. 3(b), we see that OQPSK can be interpreted as being composed of a degraded four-state

1894

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 52, NO. 11, NOVEMBER 2004

ciency, a simplied receiver can be formed by grouping sets of waveforms together based on their similarities. In particular, the are divided into eight waveforms of the I channel output four groups, and so are the eight waveforms of the Q channel . For , the th ( 0, 1, 2, 3) group consists of output and . For , when 0, 2, the waveforms th group consists of waveforms and , and when 1, 3, of waveforms and . By dening and ( 0, 1, 2, 3) as the average of the waveforms in each and , respectively, we have group for

.
Fig. 3. Alternative implementations of MIL-STD SOQPSK and OQPSK baseband signals in each symbol interval. (a) MIL-STD SOQPSK. (b) OQPSK.

trellis encoder and a memoryless signal mapper.4 The trellis of this degraded trellis encoder is exactly the one depicted in Fig. 2 , ] indicated below each branch. For OQPSK, with [ it is obvious from Fig. 3(b) that the I and Q channel-output waveform indexes depend only on their own channel-input data. Therefore, the signal mapping for OQPK is not cross-correlated, and independent I and Q channel detection is possible for OQPSK. Note that the decompositions of OQPSK and MIL-STD SOQPSK into a (degraded) trellis encoder and a memoryless mapper is important since, as will be shown in Section VI, it allows iterative decoding of the outer codes and these inherent trellis codes in their corresponding coded systems. IV. RECEIVER STRUCTURES FOR MIL-STD SOQPSK In accordance with the foregoing representation of MIL-STD SOQPSK as a TCM with four states, in this section, we present the corresponding optimum receiver structure and propose a simplied receiver structure for it. A. Optimum Receiver The optimum receiver employing a VA for maximum-likelihood sequence detection is illustrated in Fig. 4. It consists of a bank of eight matched lters (four in each of the I and Q channels) followed by a four-state trellis decoder. Note that although members of the I and Q signaling sets and (, ) do not all have equal energy, energy biases in the matched lters are not necessary when the matched-lter outputs are used in a VA. This is because the sum of the energies from allowable pairs of I and Q signals is constant, which has been explicitly veried in [20], and which comes from the fact that MIL-STD SOQPSK is a constant-envelope modulation. B. Simplied Receiver In a desire to reduce the complexity of the optimum receiver in Fig. 4 with the hope of not sacricing signicant power ef4What is meant by degraded trellis encoder is a degenerate form of such an encoder having no memory.

and , 0, Note that since 1, 2, 3, we have , , 0, 1. and ( 0, 1) are illustrated in The waveforms for and ( 0, 1) are [20], and the waveforms for and , but with opposite of the same shape as those of signs. Now we replace the waveform assignments of the group and by their corresponding average members for and both become ; and waveform, i.e., both become , and so on. Then, because of the relation between the I and Q coded bits and the BCD signal mapping in Fig. 3(a), the cross-correlation between the I and Q channel would disappear. This is because what distinguishes the is the least-signicant two waveforms in each group for for . If no distinction bit , and it is the middle bit needs to be made in each group, we can simply drop the bits and , and just use the remaining two bits , and , in each channel to specify the transmitted waveform pair ( , ), , 0, 1, 2, 3. That is

By inspecting Fig. 3(a), we see that this is equivalent to the I channel signal being chosen based only on the I encoder outputs, and the Q channel signal being chosen based only on the Q encoder outputs. Thus, the cross-correlation of the encoder outputs in choosing the I and Q waveforms disappears, and the trellis structure of the modulation decouples into two independent (I and Q) two-state trellises. The simplied Viterbi receiver corresponding to the two-state trellises is illustrated in Fig. 5. In this simplied receiver, the I and Q decisions are separately generated by individual two-state VAs using the energy-biased correlations derived from the I and Q demodulated signals, respectively. Note that, since the energy and ( 0, 2) as well as per symbol is different for and ( 0, 2), the energy biases must be set in for the matched-lter outputs, as shown in Fig. 5. In this gure, and denote the energy per symbol for and ( 0, 1), respectively. It is easily calculated that and that . Of course, the two VAs for the I and Q channels can be combined into a single four-state VA if desired. Compared with the optimum Viterbi receiver, the simplied one reduces the number of correlators by half. This simplied receiver for MIL-STD SOQPSK is very

LI AND SIMON: PERFORMANCE OF CODED OQPSK AND MIL-STD SOQPSK WITH ITERATIVE DECODING

1895

Fig. 4. Optimum receiver structure for MIL-STD SOQPSK.

similar to the one for FQPSK described in [17, Fig. 5] in both structure and complexity. In addition, it is also similar in concept to the average matched lter (AMF) receivers for continuousphase frequency-shift keying (CPFSK) [21] and CPM schemes [2]. V. PERFORMANCE COMPARISON OF UNCODED OQPSK, MIL-STD SOQPSK, AND FQPSK In this section, we rst analyze the asymptotic BER performance of MIL-STD SOQPSK and then compare the simulated BER performance of uncoded MIL-STD SOQPSK with both the optimum receiver and the simplied receiver to that of OQPSK and FQPSK. A. Asymptotic BER Performance of MIL-STD SOQPSK In this section, we obtain an expression for the minimum Euclidean distance associated with the symbol-by-symbol trellis representation of MIL-STD SOQPSK shown in Fig. 2. and the Supposing that the initial phase state is all-zero sequence is transmitted, from Fig. 2, we see that there is a path of length two that starts and ends at the same phase but differs from the all-zero sequence path. This state shortest-length error-event path passes through the phase state rst and then back to the phase state . , ] The two corresponding output waveform pairs are [

, ], respectively. The output waveform pair assoand [ ciated with each zero-input pair for initial phase state is [ , ]. Therefore, the minimum squared Euclidean distance between the shortest-length path and the all-zero sequence path is

The average signal from

energy per symbol

is obtained

Therefore, the normalized minimum squared Euclidean distance is (3) Examination of other length-two error-event paths relative to transmitted sequences other than the all-zero sequence reveals that the smallest value of normalized squared Euclidean distance

1896

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 52, NO. 11, NOVEMBER 2004

Fig. 5. Simplied Viterbi receiver for MIL-STD SOQPSK.

is still given by (3).5 Furthermore, error-event paths longer than length two correspond to larger values of squared Euclidean distance. Thus, the normalized minimum squared Euclidean distance for MIL-STD SOQPSK is given by (3). An alternative method for obtaining the result in (3) is by using the bit-by-bit CPM representation of MIL-STD SOQPSK, the details of which are given in [20]. When compared to OQPSK, which has the same normalized minimum squared Euclidean distance as binary (B)PSK, i.e., , there is a loss of 0.638 dB for MIL-STD SOQPSK. In addition, when compared with FQPSK, which is more spectrally efcient (see [4, Figs. 6 and 9]) and for which it was shown , there is an asymptotic gain of in [9] and [10] that 0.441 dB for MIL-STD SOQPSK. B. Simulation Results We have simulated the optimum receiver structure shown in Fig. 4 for uncoded MIL-STD SOQPSK, as well as the simplied receiver structure shown in Fig. 5. The numerical results are illustrated in Fig. 6. Also shown in Fig. 6 are the simulated
5In fact, we believe that SOQPSK, in general, has the uniform error property, since the Euclidean distance between two effective data sequences  and  in its CPM representation only depends on their difference sequence   [6].

= 0

BER performance of FQPSK with the optimum receiver and with a simplied receiver, which are taken from [9], [10], and [17]. Furthermore, in Fig. 6, we have also given the simulated BER performance of uncoded OQPSK, which provides a lower bound for both MIL-STD SOQPSK and FQPSK at high values. To simulate the performance of OQPSK with Viterbi decoding, the bit-interval eight-state trellis illustrated in Fig. 1 is used. In addition, we have also simulated the case where the symbol-interval trellis representation of OQPSK illustrated in Fig. 2 is used. The simulation results using these two different trellis representations of OQPSK with a Viterbi decoder are the same, and they also match the theoretical BER of BPSK and QPSK. , MIL-STD SOQPSK From Fig. 6, we see that at BER with the optimum receiver is about 0.31 dB worse than OQPSK, but is about 0.46 dB better than FQPSK with optimal reception. The simplied MIL-STD SOQPSK receiver has a performance , that is very close to the optimum receiver: at BER loss is only about 0.12 dB. For FQPSK, the perforthe mance gap between the simplied receiver and the optimum one loss at BER is roughly 0.27 dB. is bigger: the The smaller performance gap between the simplied receiver and the optimum receiver for MIL-STD SOQPSK may be due to the fact that the simplied MIL-STD SOQPSK receiver only reduces the number of matched lters in the optimum receiver

LI AND SIMON: PERFORMANCE OF CODED OQPSK AND MIL-STD SOQPSK WITH ITERATIVE DECODING

1897

Fig. 7. Recursive version of the I and Q encoders for MIL-STD SOQPSK and OQPSK. (a) MIL-STD SOQPSK. (b) OQPSK.

Fig. 6. BER performance comparison of uncoded systems.

shown that only one recursive version is allowed for each of the encoders in Fig. 3, and the recursive equivalents of these encoders are illustrated in Fig. 7. Note that if we replace the original encoders of OQPSK and MIL-STD SOQPSK, shown in Fig. 3, with their recursive equivalents, shown in Fig. 7, simulation results (not illustrated) show that the BERs of the uncoded OQPSK and MIL-STD SOQPSK are now twice those values, of the original systems shown in Fig. 6 at high similar to the phenomena observed for turbo codes and serially concatenated codes [22]. However, for the coded cases, the recursive versions of the encoders provide signicant coding gains, which will be demonstrated through a few examples in the following subsections. B. Serial Concatenation We rst consider the serially concatenated coded OQPSK and MIL-STD SOQPSK systems, similar to the coded FQPSK system illustrated in [13, Fig. 5(b)]. Specically, the input data are rst encoded by an outer encoder, interleaved, and then applied to the I and Q channels of the equivalent baseband transmitter for OQPSK or the equivalent baseband transmitter for MIL-STD SOQPSK shown in Fig. 3, where the I and Q inner encoders are replaced by their recursive counterparts in Fig. 7. After transmission over the additive White Gaussian noise (AWGN) channel, for MIL-STD SOQPSK, the received signals of the I and Q channels are passed through a bank of eight matched lters to generate a total of 16 correlator outputs, as shown in Fig. 4. For OQPSK, the I channel received signal is passed through one matched lter, and the Q channel received signal is passed through two matched lters to generate a total of six correlator outputs. These correlator outputs are then used by a four-state soft-input soft-output (SISO) [23] iterative decoder as branch metrics. For OQPSK, since there is no correlation between the I channel and the Q channel, two separate two-state iterative decoders can be used instead of a combined four-state decoder. These two decoding schemes have the same BER performance, which is veried by our simulations. For MIL-STD SOQPSK, the four-state joint I and Q channel iterative decoder must be used, since there exist correlations between these two channels. Note that in addition to the SISO module for decoding the inner code provided by OQPSK or MIL-STD SOQPSK, there is also a SISO module for decoding the outer code. To reduce the complexity of the inner and outer SISOs, we simulate the max-log versions of them, which are equivalent to modied soft-output VAs (SOVAs) [24]. The simulation results will be given in Section VI-E.

by a factor of two, while the simplied FQPSK receiver reduces it by a factor of four. VI. CODED OQPSK AND MIL-STD SOQPSK WITH ITERATIVE DECODING Since both OQPSK and MIL-STD SOQPSK have inherent trellis codes, as shown in Fig. 3, these trellis codes can be viewed as the inner code of a concatenated code in coded OQPSK and MIL-STD SOQPSK systems. As was true for the FQPSK applications [11][13], in order to realize coding gains from the concatenation of the outer code and the inherent inner code of OQPSK or MIL-STD SOQPSK, the I and Q inner encoders of the equivalent transmitters in Fig. 3 must be replaced by their recursive equivalents. Therefore, before presenting the serially and parallel concatenated coding structures for coded OQPSK and MIL-STD SOQPSK systems and introducing the simplied iterative decoding process for MIL-STD SOQPSK, we rst describe the recursive I and Q encoders for both OQPSK and MIL-STD SOQPSK. A. Recursive I and Q Encoders of OQPSK and MIL-STD SOQPSK For OQPSK and MIL-STD SOQPSK, given the original trellis of each nonrecursive I or Q encoder, the remapped recursive encoder must have a trellis for which the output bits corresponding to each transition between states remain unchanged. The only changes allowed are the input bit(s) associated with each transition. This is to guarantee that the allowable OQPSK or MIL-STD SOQPSK encoder output sequences remain unchanged so that the remapping does not change the envelope and spectral characteristics of the modulated signals [11]. Under this consideration, it can be easily

1898

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 52, NO. 11, NOVEMBER 2004

C. Parallel Concatenation We now consider a parallel concatenated coding scheme of the turbo-coding type similar to the coded FQPSK system illustrated in [13, Fig. 5(c)]. In this turbo-coding scheme, there is no explicit outer code, but rather the input bits and their interleaved versions are applied to the inherent I and Q channel encoders of OQPSK or MIL-STD SOQPSK directly. Therefore, there is no corresponding outer SISO decoder at the receiver. Instead, in each iteration, the output extrinsic information of the I and Q input bits from the inner SISO decoder(s) are fed back as reliabilities of the opposite (I to Q and Q to I) bits after appropriate interleaving and deinterleaving. At the end of the last iteration, the extrinsic information for the I input bits is combined with the deinterleaved extrinsic information for the Q input bits to produce a decision on the input information bits. D. Simplied Iterative Decoding for MIL-STD SOQPSK For both serial and parallel concatenations of coded MIL-STD SOQPSK, it is obvious that the inner SISO decoder for decoding the inherent code of MIL-STD SOQPSK can be replaced with two separate and simplied SISO decoders, using the trellis diagrams of the simplied I and Q encoders shown in Fig. 5. Now a joint SISO decoder for decoding the I and Q input bits of the inherent encoder in MIL-STD SOQPSK is not necessary, due to the decoupling of the I and Q channel transmissions. Consequently, the number of matched lters required at the receiver can be reduced by half. It will be shown in the next subsection that although there is a noticeable performance loss due to the simplication, this reduced-complexity scheme has a performance comparable to that of coded FQPSK with iterative decoding using the full-blown matched-lter conguration [10], and it is superior to coded FQPSK with iterative decoding using a simplied receiver [11][13], [17].6 E. Simulation Results In this section, we present simulation results for serially and parallel concatenated coded OQPSK and MIL-STD SOQPSK with iterative decoding. Also presented are simulation results of the simplied iterative decoding for MIL-STD SOQPSK, as described in the previous subsection. In addition, for comparison, we have simulated in each case the performance of coded FQPSK with iterative decoding using full-blown matched lters [10], as well as using a simplied receiver structure [11][13], [17]. In our simulations, for both parallel and serially concatenated cases, no termination bits are added anywhere and the decoder state metrics are uniformly initialized in each data block. For serial concatenation, we have investigated two different outer codes. The rst one is a recursive rate-1/2 optimum four-state . The convolutional code with minimum free distance generator polynomials and of its nonrecursive equivalent have an octal representation of 5 and 7, respectively. The second outer code considered is a higher rate (i.e., 3/4) recursive code, obtained by treating the the rst one as a rate-3/6 code, and then
6Note, however, that MIL-STD SOQPSK is less bandwidth-efcient than FQPSK.

Fig. 8. BER performance of coded systems with serial and parallel concatenations.

puncturing two of the output bits. The minimum free distance . Compared to the rate-1/2 code, this of this code is code is more bandwidth-efcient,7 but we will see that it is less power-efcient. In our simulations, the number of iterations is , and the interleaver block size is 2048 bits (1024 information bits) for the rate-1/2 outer code. For the rate-3/4 1364 outer code, the interleaver block size is chosen to be bits (1364 (3/4) = 1023 information bits). Note that since is small, for better performance, we have scaled the extrinsic information from the inner SISO(s) and the outer SISO by a factor of 0.75 for the rate-1/2 outer code [25]. For the rate-3/4 outer code, we have scaled the extrinsic information from the inner SISO(s) to the outer SISO by a factor of 0.7, while leaving the extrinsic information from the outer SISO to the inner SISO(s) unchanged. For parallel concatenation,8 the number of iterations is , and the interleaver block size is 2048 bits (1024 information bits). In this case, no scaling factor is applied to the extrinsic information from the inner SISO decoder. Fig. 8 shows the BER performance of the three parallel concatenated coding systems as well as that of the three serially concatenated coding systems with both the rate-1/2 outer code and the rate-3/4 outer code. From Fig. 8 we see that, for serial concatenation with the rate-1/2 outer code, coded OQPSK has the best performance, as expected, and its performance is -differvery similar to that of serially concatenated coded ential QPSK (DQPSK) with the same rate-1/2 outer code [26]. The performance of coded MIL-STD SOQPSK is only slightly worse than that of coded OQPSK, and it is noticeably better than , the required gap that of coded FQPSK. At BER between coded OQPSK and coded MIL-STD SOQPSK is less
7The rate-1/2 code gives an overall rate of 1 bit/symbol for the serially concatenated system, while the rate-3/4 code gives an overall rate of 1.5 bits/symbol. 8The overall rate of the parallel concatenated coding scheme is 1 bit/symbol.

LI AND SIMON: PERFORMANCE OF CODED OQPSK AND MIL-STD SOQPSK WITH ITERATIVE DECODING

1899

than 0.02 dB, while it is about 0.09 dB between coded OQPSK and coded FQPSK. The simplied iterative decoding of coded MIL-STD SOQPSK is about 0.1 dB worse than the nonsimpli, and it is very close to that of ed decoding at BER coded FQPSK with iterative decoding using full-blown matched lters. The simplied decoding of coded FQPSK, however, is about 0.15 dB worse than the nonsimplied decoding at BER . Compared with the uncoded cases shown in Fig. 6, , the coding gains for OQPSK, MIL-STD SOat BER QPSK, and FQPSK are 7.05, 7.35, and 7.78 dB, respectively, and the coding gains for the simplied receivers of MIL-STD SOQPSK and FQPSK are 7.43 and 7.88 dB, respectively. Therefore, of the three modulation schemes, the more bandwidth-efcient a scheme is, the greater the coding gain. For serial concatenation with the rate-3/4 outer code, the relations of the ve BER curves are very similar to those for the rate-1/2 outer code, except that now the performance of the simplied decoding of coded MIL-STD SOQPSK is almost indistinguishable from that of coded FQPSK with nonsimplied decoding. In addition, unlike the rate-1/2 outer-code case, the increases, BERs are not decreasing so dramatically as ranges. This is because the interleaver especially at high size is smaller in the rate-3/4 outer-code case, and the rate-3/4 code has a smaller free distance than the rate-1/2 code [22]. Still, with this rate-3/4 outer code, there are signicant coding gains , for when compared with the uncoded case. At BER OQPSK, MIL-STD SOQPSK, and FQPSK, they are 5.77, 6.03, and 6.41 dB, respectively, and for the simplied decoding of MIL-STD SOQPSK and FQPSK, they are 6.10 and 6.45 dB, respectively. Finally, we observe from Fig. 8 that, with the simple parallel concatenated coding scheme, the relations of the ve BER curves are very similar to those of their serial concatenation counterparts, except that now the BERs are decreasing much increases. In fact, the ve BER curves start slower as to show leveling off even before the BER reaches , which is similar to the error oor scenario of parallel concatenated convolutional codes [22]. For this simple parallel concatenated coding scheme, when compared with the uncoded systems, are 4.76, 5.03, and 5.42 dB the coding gains at BER for OQPSK, MIL-STD SOQPSK, and FQPSK, respectively, and they are 5.08 and 5.46 dB for the simplied decoding of MIL-STD SOQPSK and FQPSK, respectively. Although these coding gains are not as big as those obtained with serial concatenation of the rate-1/2 outer code, or even the higher rate (i.e., 3/4) outer code, the iterative decoding complexity of the parallel concatenated coding scheme is much lower than its serial concatenation counterparts due to the lack of necessity of an outer SISO decoder. Therefore, tradeoffs must be made between receiver complexity, coding gain, and bandwidth efciency when designing coded OQPSK, MIL-STD SOQPSK, and FQPSK systems. VII. CONCLUSION We have shown how both OQPSK and MIL-STD SOQPSK can be decomposed into a (degraded) trellis encoder and a memoryless mapper. When concatenated with an outer code, coded

OQPSK with iterative decoding provides a lower bound to the performance of coded MIL-STD SOQPSK and FQPSK. For MIL-STD SOQPSK, we have analyzed its asymptotic BER performance, presented the optimum receiver structure, and proposed a simplied receiver. The simplied receiver maintains good performance with reduced complexity. Simulation results show that the performance of coded MIL-STD SOQPSK comes very close to that of coded OQPSK, and is noticeably better than that of coded FQPSK, though MIL-STD SOQPSK is less bandwidth-efcient than FQPSK. When compared with their uncoded systems, there are signicant coding gains for both coded OQPSK and MIL-STD SOQPSK applying iterative decoding to either the parallel concatenated coding scheme or the serial one, even when very simple outer codes are used. ACKNOWLEDGMENT The authors would like to thank Dr. D. Divsalar and D. Lee for many helpful discussions. REFERENCES
[1] M. J. Dapper and T. J. Hill, SBPSK: A robust bandwidth-efcient modulation for hard-limited channels, in Proc. IEEE Military Commun. Conf., Los Angeles, CA, Oct. 1984. [2] J. B. Anderson, Digital Phase Modulation. New York: Plenum, 1986. [3] T. J. Hill, An enhanced, constant envelope, interoperable shaped offset QPSK (SOQPSK) waveform for improved spectral efciency, in Proc. Int. Telemetering Conf., San Diego, CA, Oct. 2000, pp. 127136. [4] , A nonproprietary, constant envelope, variant of shaped offset QPSK (SOQPSK) for improved spectral containment and detection efciency, in Proc. IEEE Military Commun. Conf., vol. 1, Los Angeles, CA, Oct. 2000, pp. 347352. [5] P. S. K. Leung and K. Feher, F-QPSKA superior modulation technique for mobile and personal communications, IEEE Trans. Broadcast., vol. 39, pp. 288294, June 1993. [6] M. Geoghegan, Implementation and performance results for trellis detection of SOQPSK, in Proc. Int. Telemetering Conf., Las Vegas, NV, Oct. 2001, pp. 531540. , Bandwidth and power efciency trade-offs of SOQPSK, in Proc. [7] Int. Telemetering Conf., San Diego, CA, Oct. 2002, pp. 780789. [8] M. K. Simon and T.-Y. Yan, Cross-Correlated Trellis-Coded Quadrature Modulation, U.S. patent led Oct. 1999. [9] M. K. Simon and T.-Y. Yan. (1999, May) Performance evaluation and interpretation of unltered Feher-patented quadrature phase-shift keying (FQPSK). Telecommunications and Mission Operations Prog. Rep. [Online]. Available: http://tmo.jpl.nasa.gov/tmo/progress_report/42-137/137C.pdf [10] , Unltering Feher-patented quadrature phase-shift keying (FQPSK): Another interpretation and further enhancements: Parts 1, 2, Appl. Microwave Wireless Mag., vol. 12, p. 7696/100105, Feb./Mar. 2000. [11] M. K. Simon and D. Divsalar. (2001, May) A reduced-complexity, highly power-/bandwidth-efcient coded Feher-patented quadrature phase-shift keying system with iterative decoding. Telecommunications and Mission Operations Prog. Rep. [Online]. Available: http://tmo.jpl.nasa.gov/tmo/progress_report/42-145/145A.pdf [12] , (2001, Aug.) Further results on a reduced-complexity, highly power-/bandwidth-efcient coded Feher-patented quadrature phaseshift keying system with iterative decoding. Interplanetary Network Prog. Rep. [Online]. Available: http://tmo.jpl.nasa.gov/tmo/progress_report/42-146/146I.pdf [13] , A reduced complexity highly power/bandwidth efcient coded FQPSK system with iterative decoding, in Proc. IEEE Int. Conf. Communications, vol. 7, Helsinki, Finland, June 2001, pp. 22042210. [14] P. Moqvist and T. M. Aulin, Serially concatenated continuous phase modulation with iterative decoding, IEEE Trans. Commun., vol. 49, pp. 19011915, Nov. 2001. [15] K. R. Narayanan and G. L. Stber, Performance of trellis-coded CPM with iterative demodulation and decoding, IEEE Trans. Commun., vol. 49, pp. 676687, Apr. 2001.

1900

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 52, NO. 11, NOVEMBER 2004

[16] M. R. Shane and R. D. Wesel, Reduced complexity iterative demodulation and decoding of serial concatenated continuous phase modulation, in Proc. IEEE Int. Conf. Commununications, vol. 3, Apr. 2002, pp. 16721676. [17] D. Lee, M. K. Simon, and T.-Y. Yan, Enhanced performance of FQPSK-B receiver based on trellis-coded Viterbi demodulation, in Proc. Int. Telemetering Conf., San Diego, CA, Oct. 2000, pp. 631640. [18] M. K. Simon, Bandwidth-Efcient Digital Modulation With Application to Deep-Space Communication. New York: Wiley, 2003. [19] M. K. Simon and L. Li. (2003, Aug.) A cross-correlated trellis-coded quadrature modulation representation of MIL-STD shaped offset quadrature phase-shift keying. Interplanetary Network Prog. Rep. [Online]. Available: http://ipnpr.jpl.nasa.gov/progress_report/42-154/154J.pdf [20] L. Li and M. K. Simon. (2004, Feb.) Performance of coded offset quadrature phase-shift keying (OQPSK) and MIL-STD shaped OQPSK (SOQPSK) with iterative decoding. Interplanetary Network Prog. Rep. [Online]. Available: http://ipnpr.jpl.nasa.gov/progress_report/42-156/156A.pdf [21] W. Osborne and M. B. Luntz, Coherent and noncoherent detection of CPFSK, IEEE Trans. Commun., vol. COM-22, pp. 10231034, Aug. 1974. [22] S. Benedetto, D. Divsalar, G. Montorsi, and F. Pollara, Serial concatenation of interleaved codes: Performance analysis, design, and iterative decoding, IEEE Trans. Inform. Theory, vol. 44, pp. 909926, May 1998. , A soft-input soft-output APP module for iterative decoding of [23] concatenated codes, IEEE Commun. Lett., vol. 1, pp. 2224, Jan. 1997. [24] M. P. C. Fossorier, F. Burkert, S. Lin, and J. Hagenauer, On the equivalence between SOVA and max-log-MAP decodings, IEEE Commun. Lett., vol. 2, pp. 137139, May 1998. [25] D. Kim, T. Kwon, J. R. Choi, and J. J. Kong, A modied two-step SOVA-based turbo decoder with a xed scaling factor, in Proc. IEEE Int. Symp. Circuits and Systems, vol. 4, Geneva, Switzerland, May 2000, pp. 3740. [26] K. R. Narayanan and G. L. Stber, A serial concatenation approach to iterative demodulation and decoding, IEEE Trans. Commun., vol. 47, pp. 956961, July 1999.

Lifang Li received the B.S. and M.S. degrees from Tsinghua University, Beijing, China, in 1993 and 1996, respectively, and the Ph.D. degree from the California Institute of Technology, Pasadena, in 2000, all in electrical engineering. From February 2000 to April 2000, she worked as a postdoctoral scholar in the Advanced Communications Systems Concepts Group, Jet Propulsion Laboratory (JPL), Pasadena, CA. In May 2000, she joined the Exeter Group, Inc., Los Angeles, CA. Since November 2002, she has been a Member of the Technical Staff with the Information Processing Group, JPL.

Marvin K. Simon (S60M66SM75F78) is currently a Principal Scientist at the Jet Propulsion Laboratory (JPL), California Institute of Technology (Caltech), Pasadena, where for the last 36 years he has performed research as applied to the design of NASAs deep-space and near-earth missions, and which has resulted in the issuance of nine U.S. patents, 25 NASA Tech Briefs, and four NASA Space Act awards. He is known as an internationally acclaimed authority on the subject of digital communications with particular emphasis in the disciplines of modulation and demodulation, synchronization techniques for space, satellite, and radio communications, trellis-coded modulation, spread spectrum and multiple access communications, and communication over fading channels. He has also held a joint appointment with the Electrical Engineering Department at Caltech. He has published over 195 papers on the above subjects and is coauthor of 10 textbooks, including Telecommunication Systems Engineering (Englewood Cliffs, NJ: Prentice-Hall, 1973, and New York: Dover Press, 1991), Phase-Locked Loops and Their Application (New York: IEEE Press, 1978), Spread Spectrum Communications, Vols. I, II, and III (Computer Science Press, 1984 and New York: McGraw-Hill, 1994), An Introduction to Trellis Coded Modulation with Applications (MacMillan, 1991), Digital Communication Techniques: Vol. I (Englewood Cliffs, NJ: Prentice-Hall, 1994) and Digital Communication Over Fading Channels: A Unied Approach to Performance Analysis (New York: Wiley, 2000, 2nd ed. available Nov. 2004), Probability Distributions Involving Gaussian Random VariablesA Handbook for Engineers and Scientists (Norwell, MA: Kluwer, 2002) and Bandwidth-Efcient Digital Modulation with Application to Deep-Space Communication (New York: Wiley, 2003). His work has also appeared in the textbook Deep Space Telecommunication Systems Engineering (New York: Plenum, 1984), and he is coauthor of a chapter entitled Spread Spectrum Communications in the Mobile Communications Handbook (Boca Raton, FL: CRC Press, 1995), Communications Handbook (Boca Raton, FL: CRC Press, 1997), and the Electrical Engineering Handbook (Boca Raton, FL: CRC Press, 1997). His work has also appeared in the textbook Deep Space Telecommunication Systems Engineering (New York: Plenum, 1984). Dr. Simon is the corecipient of the 1986 Prize Paper Award in Communications of the IEEE Vehicular Technology Society and the 1999 Prize Paper Award of the IEEE Vehicular Technology Conference (VTC99-Fall), Amsterdam, The Netherlands. He is a Fellow of the IAE, and winner of a NASA Exceptional Service Medal, a NASA Exception Engineering Achievement Medal, the IEEE Edwin H. Armstrong Achievement Award, and most recently, the IEEE Millennium Medal.