Superposed Signaling Option for Bandwidth Efficient Wireless LANs

Superposed Signaling Option for Bandwidth Efficient Wireless LAs Thomas Deckert, Wolfgang Rave, and Gerhard Fettweis Vodafone Chair Mobile Communications Systems Dresden University of Technology, 01062 Dresden, Germany {deckert,rave,fettweis}@ifn.et.tu-dresden.de Abstract One key challenge for increasing the throughput of future wireless local area networks at higher layers is matching the capabilities of the physical (PHY) layer with the need of exchanging small control messages at these higher layers. The bottleneck in this regard lies at the medium access control (MAC) layer that may waste considerable channel resources in the transmission of these small packets. In this contribution we introduce superposed signaling, a PHY layer extension that opens up a new way to map small packets to the channel while reducing inefficient use of the medium. This scheme provides a low-rate signaling sublayer at the PHY level in addition to a high-rate data sublayer. These sublayers share the channel by a variant of multi-carrier code-division multiple access (MC-CDMA). We investigate the design of receivers that are capable of detecting the signals of both sublayers, and focus on a structure employing successive interference cancellation (SIC). Our main interest is in identifying the operation modes in terms of per-user powers in which it is possible to use these receivers. Furthermore, it will be shown that the scheme is appropriate for sublayer-synchronous as well as asynchronous scenarios. Keywords Superposed signaling, multi-carrier spread spectrum,, successive interference cancellation. I. ITRODUCTIO Recent years have seen the rise of wireless local area network (WLA) services. Current state-of-the-art standards such as IEEE 802.11a/g, or HiperLA/2 employ orthogonal frequency division multiplexing () to support physical (PHY) layer data rates of up to 54 Mbps ([1]), and already there are visions of future based 4G systems capable of peak data rates of up to 1 Gbps ([2]). The role of the medium access control (MAC) in efficiently using the PHY data rates on the higher layers has been widely acknowledged ([3]). Here we identify a yet unaddressed problem of using existing MAC protocols to map typical high-rate and lowrate traffic to future -based PHYs. We introduce a novel way to multiplex such traffic on the PHY layer to significantly improve higher layer throughput. The packets delivered to the MAC can be large, yet invariably there is a high proportion of rather small packets. Indeed, about 60% of all TCP packets are smaller than 50 bytes ([4]). Such packets mostly carry control signaling information of the higher layers. Currently, the MAC protocols essentially time-multiplex all packets. Together with an -based PHY this may soon be very inefficient. One symbol corresponds to several, e. g., quadrature amplitude modulated (QAM) symbols that are transmitted in one go. This way, in 802.11a there are up to 27 bytes per symbol at the highest transmission rate (54 Mbps, with symbol duration equal to 4 µs). In future systems this number may easily become several hundred bytes (at 1 Gbps and symbol duration as in 802.11a: 500 bytes). Thus, reserving the full channel resources for the transmission of one small signaling packet will enormously waste bandwidth. Here, we propose to use multi-carrier spread-spectrum () techniques for superimposing control signaling of one user with regular data of another to exploit the scarce bandwidth efficiently. On the PHY layer we have to deal with multi-access interference (MAI), and we will focus on investigating the operating modes of single-user as well as multi-user receivers. Our goal is to design the system such that the receiver can be simple while keeping the transmission powers near their single-user limits. Section II will introduce the PHY layer system concept of superposed signaling. Then we will examine analytically, and validate numerically, the principal behavior of the single-user receiver in Section III, and that of the multi-user receiver in Section IV. Finally, the influence of user asynchrony on the performance of the system, and our analysis, is the focus of Section V. II. SYSTEM COCEPT: SUPERPOSED SIGALIG We envision a signaling sublayer of the -based PHY layer to carry the control frames of one user and a data sublayer to transport larger data packets of another user. The system uses the same set of subcarriers to carry both sublayers simultaneously. For high-rate transmission of bulky user data (coded) is employed while the low-rate signaling sublayer is superimposed on the signal via a technique known as multi-carrier code-division multiple access (MC-CDMA) ([5]). In the latter scheme a QAM symbol is multiplied with a length- spreading sequence and each chip of that sequence modulates one subcarrier, which amounts to spreading directly in the frequency domain. The choice of MC-CDMA for the signaling sublayer is motivated by two reasons. Firstly, the spreading process results in rather low power per subcarrier, i. e., low interference to the signal. This effect is more pronounced the larger the number of subcarriers is. Secondly, while the data rate of a single MC-CDMA user is low (which is fine for the signaling sublayer) the scheme exploits the full frequency diversity thus making transmission more reliable. While there is one user in the data sublayer only there may be several users in the signaling sublayer. For simplicity but without loss of generality we restrict ourselves to one user each in the two sublayers. The users are allowed to access the medium asynchronously. Transmitting the two signals in that manner causes MAI at the receiver. If the interference levels can be held low enough simple single-user detection as in Fig. 1 may be used. However, as explained in Section III, this requires large transmission powers, and might not be possible at all. Alternatively, we propose a multi-user receiver that employs successive interference cancellation (SIC) to suppress the MAI sufficiently. Fig. 2 shows the structure of the SIC receiver. III. OPERATIO MODES OF SIGLE-USER RECEPTIO In this section we will look at the signal-to-noise ratio (SR) requirements for each stream such that the single-user receiver of Fig. 1 can detect both sublayers with their targeted error rate. Single-user detectors regard the received signal as the sum of the desired signal and additive white Gaussian noise (AWG) without treating MAI specifically. However, if the interference is

d Channel 1 ˆd c Channel 2 oise ĉ reception + Fig. 1. Single-user receiver structure. Each sublayer is detected independently of the other. d c Channel 1 Channel 2 s(t) ˆd Synthesis ŝ(t) - ĉ SR + M reception oise SR + O Fig. 2. Multi-user receiver with successive interference cancellation (SIC receiver). The signal is detected first using a single-user receiver. The signal is detected after the contribution of the sublayer to the received signal has been cancelled. Fig. 3. Single-user receiver: SRs required for stream reception. Shaded areas are appropriate combinations for either or detection. In the overlapping region both streams can be detected with the required performance. approximately white it just results in a larger effective AWG level. Then the error probability is directly related to the ratio of the desired signal power and the combined interference and noise powers, SIR. To achieve the required error rates we need to ensure that the SIRs of the and streams, SIR O and SIR M, are above the thresholds, and, associated with these error rates, i. e., SIR O = γo, 1 + / (1) SIR M =. 1 + (2) Here and are the signal-to-noise ratios of the and streams. ote that SIR O and SIR M correspond to QAM symbols at the demodulator input. Thus, in transmission SIR O is the SIR of an individual subcarrier while in transmission SIR M is the SIR after despreading, i. e., corresponds to all subcarriers. The thresholds and can be determined from the single-user reference performance of the and stream, respectively. For the single-user receiver structure to work we need to choose and to satisfy both SIR criteria. Rewriting (1), (2), given we can bound by SRO (1 + ) 1. (3) Analogously, given the appropriate range of is 1 + SRM SRM 1. (4) Intuitively, relation (3) demands that the stream is strong enough to cover its single-user requirement and offset the additional interference while at the same time being weak enough such that the stream is not disturbed too much. Relation (4) states the same requirement in terms of. Fig. 3 illustrates in a SR plane (3) and (4). The lines separate the regions where the SIRs are sufficient from those where the SIRs are too low for the required detection quality. If they intersect for some SR + O > and SR + M > then there is an operation region where both SIR criteria are satisfied. Thus, we can employ the single-user receiver whenever such an intersection occurs, which happens for >, (5) and at SR + + O = >, (6) SR + M = (1 + SR + O). (7) Observe that the amount of additionally required power is higher than in individual single-user transmission of each stream. This is especially pronounced in the sublayer. We validate our theoretical analysis via the following example. Consider a system with = 512 and 100 MHz bandwidth at a carrier frequency of 5 GHz that uses the single-user receiver structure. The channel is modeled as block-fading with power delay profile according to the HiperLA/2 channel model A [6]. The cyclic prefix is such that there is flat fading on each subcarrier. All information bits are convolutionally coded with code rate 1/2. The length of a code word is 576 bytes in the sublayer, and 12 bytes in the sublayer. The stream carries 16-QAM symbols, the stream uses QPSK. The sublayer single-user detectors include demapping, soft demodulation and Viterbi decoding. Each detector is perfectly synchronized to its corresponding transmitter and has full channel knowledge. Furthermore, the target bit error rates (BER) for system design are specified as 2 () and 3 (). We determined the single-user SRs to achieve the target error rates to be = 14 db and = 4 db. (In the equations all SR terms correspond to the linear scale while we give their values in db.) We can apply single-user receivers since 10 log 10 = 27 db > 14 db + 4 db. The minimum required SRs are SR + O = 14.6 db, SR + M = 18.7 db. If we choose SRO = 15 db then needs to be between 19.1 db and 21.3 db (see (3)). Fig. 4 shows the BERs of both sublayers for varying signal strength in that scenario. This corresponds to a cross-section through the power plane along the dotted line in Fig. 3. For low the signal is practically free of MAI and can be detected with single-user BER = 3 10 5 while the stream cannot be detected at all. Increasing leads to higher MAI levels in the layer and its BER slowly grows, but the SIR M increases as well thus improving the BER of the stream. For values > 22 db the behavior of the layers is reversed with the stream being more reliable and the signal suffering severely from MAI. The operating range of the receiver where both streams satisfy

BIT ERROR RATE ( = 15dB) 10 5 16 18 20 22 24 26 Fig. 4. Single-user receiver performance: Bit error rates of and streams over the SR of the stream. The SR of the stream is 15 db. The operating range is indicated by the dashed lines. reception + (with SIC) reception Fig. 5. SIC receiver: SRs required for stream reception. Shaded areas are appropriate combinations for either or detection. With SIC both streams can be properly received for combinations in the dash-dotted pie. their BER requirements is indicated by the dashed lines in Fig. 4. From the figure we see that 19.3 db 21.7 db which essentially confirms the theoretical range obtained above. This close correspondence of theory and experiment suggests that the interference can indeed be considered Gaussian, and that the analytical results based on this assumption are valid. IV. OPERATIO MODES OF THE SIC RECEIVER In this section we look at the SR requirements to operate the SIC receiver. This receiver works even if the single-user receiver cannot be employed, i. e., whenever <, and comes with just a moderate increase in overall receiver complexity. SIC is a multi-user detection technique most suited for situations where some signals can be detected more easily than others. This is exactly what we have in our case: Assume that both layers transmit with their single-user powers, i. e., and (and ). Then the detector is much less impaired by MAI than the detector as the stream transmits just one QAM symbol per symbol and spreads its power over the entire bandwidth (see (1), (2)). Thus, we start with detecting the stream as this is robust to the interference, and cancel its effect on the received signal such that the input of the detector is essentially MAI free (see Fig. 2). Practically, errors in the detection and cancellation of the signal will lead to some residual interference power. To improve performance the process may be reiterated. For limited complexity we complete reception after one such cycle. We will see that this is sufficient to obtain near single-user performance for both sublayers. To ensure reliable detection in the presence of an signal received with we choose according to (4) as (1 + /). Thus, the operating point (, ) lies within the dark shaded area in Fig. 5. Since the detector works with reduced (ideally, no) interference we can set its SR simply to meet its single-user requirements, i. e.,. Then the operating point is on or above the horizontal dashed-dotted line in Fig. 5. Of course, there will be some impairment of the detection by residual interference. This is no problem as long as the factor limiting performance is the AWG rather than the MAI. On the other hand, if the performance is degraded by the MAI we might either run additional SIC iterations or lower the error probability by increasing. Here we take the latter approach to limit the receiver complexity. The goal is to determine when an additional constraint on is needed and how to quantify it. To that end assume completely synchronous streams and consider the detection of an individual (QAM) symbol c from the stream. Let the vector d denote the collection of QAM symbols carried in the symbol of the stream corresponding to c. Furthermore, let us define an symbol error to be the event that at least one of the QAM symbols in d is detected wrongly, and let us denote its probability by S. If no error occurred in the detection of d, the receiver attains single-user performance with a QAM symbol error rate Q SU. Otherwise, the detection of c is impaired by residual interference which leads to a QAM symbol error rate Q I > Q SU. Thus, the probability of wrongly detecting a symbol c of the stream is Q = Q SU (1 S) + Q I S. (8) A loose upper bound of (8) is Q Q SU + Q I S. Thus, as long as Q SU > S we have single-user (i. e., noise-limited) performance, Q Q SU. If Q SU S the receiver quality is determined by the interference through Q Q I S. In conclusion the conditions the SRs of the and sublayers have to satisfy to ensure proper reception of both streams are ( SRO 1 min{, γ O } 1 + SRM, ) (9), (10) where γ O corresponds to the SR required to achieve a single-user symbol error rate of S < Q SU, and, are again the single-user SR values associated with the target error rates of the and streams, respectively. With these choices the operating point of the system is within the dashed-dotted pie in Fig. 5. ote that is constrained just by a lower bound while acceptable values for are within a finite interval. The operating range of the receiver with respect to can be influenced by the SR of the stream and the number of subcarriers. How large the operating range should be is a matter of how much may vary which depends on channel fading conditions and power control. From (10) we see that for each db more in we get an enlargement of the operating range for of roughly one db. Compared to the single-user receiver (if it is realizable), the SIC receiver requires much less SR in the stream, and also less SR in the stream. (The easy calculation of that fact

: SYMBOL ERROR RATE : QAM SYMBOL ERROR RATE Operating Range Target Error Rate, SIGLE USER (MAI from ) (residual MAI from ) 5 10 15 20 25 30 35 40 45 Fig. 6. SIC receiver performance: symbol error rate and QAM symbol error rate over the SR of the stream. The SR of the stream is fixed. : SYMBOL ERROR RATE : QAM SYMBOL ERROR RATE, SIGLE USER (MAI from ) (residual MAI from ) 20 10 20 30 40 Fig. 7. SIC receiver performance: symbol error rate and QAM symbol error rate over the SR of the stream. The SR of the stream is fixed. has been omitted due to lack of space.) Thus, application of the SIC receiver helps saving transmission power. A final note on the order in which the streams are detected: It is of course possible and has been studied by the authors to reverse the detection sequence. The principal behavior of the system is retained, and the analytical treatment is completely analogous; only the roles of the and streams are interchanged. However, detecting the stream before the stream has one big advantage that makes this scheme appealing to the design of future WLA standards: The power per subcarrier that the stream requires in the presented scheme is much lower than that of the stream. Thus, if used, the signaling sublayer will create only a low interference level to other -based transmission. This may help in designing backward compatible transmission modes for advanced systems for communication with legacy devices. Reversing the detection sequence will necessitate much higher powers per subcarrier in the layer, so that scheme will be harder to combine with existing systems. ow let us verify our analysis by examining a system employing the SIC receiver. Similar to IEEE 802.11a, the system bandwidth is 20 MHz, and = 64 subcarriers are used. The stream consists of 64-QAM symbols, the sublayer transmits 16-QAM symbols. The SIC receiver uses just one iteration as considered in the analysis as well. All other system parameters are identical to those considered in Section III. For this system we determined the single-user SRs to achieve the target error rates to be = 22.8 db and = 13 db. For the stream we choose = 24.3 db such that SIR O > at = with a very low margin of 0.3 db according to (1). The maximum that we can tolerate is that for which SIR O =, and is 14.2 db in our case (see (10)). Thus the operating range of the system is 13 db 14.2 db at = 24.3 db. This rather small interval is due to our stringent system design. The target bit error rates correspond to both, the symbol error rate of the stream as well as the QAM symbol error rate of the stream, being 1. In the following, we first fix the power and vary, i. e., we go through the power plane along the dotted line in Fig. 5. Then, we take the path along the dashed line by increasing at constant. Fig. 6 shows the symbol error rate (S) and the QAM symbol error rate (Q) of the stream over. As is increased the performance degrades into the worst case behavior due to growing MAI. However, as long as < 14 db the resulting SIR (see (1)) is high enough, SIR O > 22.8 db, such that the target error rate can be achieved. At the same time the error rate is that of a single user up to about 15 db. But starting at about 13 db we have Q SU S, and according to (8) the influence of the residual interference becomes the determining factor for the performance. However, in the range of = 13 db to 15 db the error rate is still close to the single-user bound. It can be concluded that for these the influence of the residual MAI on the symbol error rate is weak, i. e., Q I Q SU. At about 15 db the residual interference becomes the limiting factor, i. e., Q SU Q I. Since the signal detection worsens dramatically with increasing further Q Q I S grows large as well. Ultimately, the signal will get strong enough such that it can be detected more reliable again even in the presence of the worst case interference (S = 1), such that Q Q I decreases. ote that for values as high as this the signal may be detected without any prior interference cancellation as it can easily accommodate for the additional noise, and it would make sense to reverse the order of detection. The target error rate is achieved for between 13 db and 23 db. Together with the above result, < 14 db for reliable detection, this confirms the theoretical operating range of the system. ow we keep = = 13 db fixed and vary. The additional SR requirement for the stream due to the interference is (1+ /) 1.2 db. Consequently, the minimum is 22.8 db + 1.2 db = 24 db. The error rates corresponding to this case are shown in Fig. 7. With increasing the performance is improved. The symbol error rate follows the corresponding single-user case with the predetermined SR gap of 1.2 db and falls below the target value 1 around = 24 db as predicted. The error rate starts at the single-user value for very low, i. e., for practically MAI free conditions. As slowly grows and the error rate remains high, the MAI level in the detector increases considerably. The error rate is high and governed by the second term in (8), Q Q I S. At about 22 db the error rate is low enough such that the detection stage can be regarded as MAI free again, and the error rate settles at its single-user value. Again, note the close correspondence between the analysis and the numerical examples. V. IFLUECE OF USER ASYCHROY Finally, we address the implications of the and users being asynchronous, i. e., of having offsets in the carrier

: SYMBOL ERROR RATE : QAM SYMBOL ERROR RATE = 36 db, ε = 0.1 (0.5) = 16 db, ε = 0.1 (0.5) = 13 db, ε = 0.1 (0.5) = 16 db, ε = 0.1 = 16 db, ε = 0.5 0 0.1 0.2 0.3 0.4 0.5 USER TIME OFFSET / SYMBOL DURATIO Fig. 8. SIC receiver performance if the and users are asynchronous: symbol error rate and QAM symbol error rate over the time offset between the users. The time offset is normalized with respect to the symbol duration. Curves parameterized by and user frequency offset. The SR of the stream is 22.8 db. frequency or the timing of the two signals. The key result is that these effects do not impair the performance of the SIC receiver in most practical cases. Thus, the previous discussion applies to these asynchronous scenarios as well. We consider the system corresponding to Fig. 6 in Section IV that uses a SIC receiver to separate the stream with fixed power ( = 22.8 db) from the stream whose power varies. Fig. 8 shows the symbol error rates for different time offsets between the two streams relative to the duration of an symbol. The curves are parameterized by the carrier frequency offset normalized to the subcarrier spacing, ɛ = 0.1 and 0.5, and the power of the stream, i. e., = 13 db, 16 db and 36 db. Apart from the performance for = 13 db the curves for ɛ = 0.5 are nearly identical to those for ɛ = 0.1. Therefore, they are omitted. This figure shows clearly that the performance does not depend on the timing and frequency offsets between the users signals. As for the performance, there is a dependency on user asynchrony for = 16 db and 36 db with the frequency offset having more influence for 16 db and the timing offset being the significant parameter for 36 db. Observe that these two cases correspond to operating the system outside its target region where the signal cannot be detected reliably and the performance depends mainly on the residual interference, i. e., Q Q IS (see (8) and Fig. 6). An intuitive explanation for this behavior is the following: First consider the detection of the stream (see Fig. 2). In the user synchronous case a specific QAM symbol d in the i-th symbol experiences interference from that chip in the signal that is transmitted on the same subcarrier as d in the i-th symbol of the stream. If the user synchrony is lost so is the orthogonality of the subcarriers of the signal as seen by the detector. (ote that the orthogonality of the subcarriers of the signal is retained in the detection stage as long as we have perfect synchronization of the detector and transmitter of the stream.) ow the interference to the QAM symbol d is composed of the contributions of many chips from the signal both, from other subcarriers and from earlier or later symbols of the stream. However, the average interference power cannot change. Thus, the performance of the stream detector will not change. In Section IV we considered two main cases for detecting the stream. In the first one we have sufficient suppression of the interference (in our example this corresponds to = 13 db). As the detector enjoys near singleuser conditions any user asynchrony is rendered irrelevant. This is confirmed by the figure. In the second case the residual interference determines the performance of the detector. The structure and statistical properties of that interference are not readily apparent and subject to further research. However, they seem to depend on the time and frequency offsets between the two streams thereby influencing the detection. ote that while the degradation in the QAM symbol error rate at = 16 db and ɛ = 0.1 from 7 to 1 is significant this is still one order of magnitude lower than the target error rate for which the system was designed (see Section IV). Also, at this value of we are outside the target operating region. Furthermore, in the case of = 36 db a time offset between the users actually helps improve performance. However, even with this effect neither of the two streams can be detected properly. Thus, we conclude that within the practically interesting operating region user asynchrony does not change the behavior of the system. VI. COCLUSIO We have presented the concept of superposed signaling, a physical layer extension for based high-rate communications systems. This scheme aims to reduce the inefficient utilization of transmission bandwidth if short signaling packets carrying control data of higher layers are mapped to the channel by the MAC layer. A MAC protocol applied on top of a physical layer with superposed signaling can transmit these small packets in the signaling sublayer while leaving the conventional high-rate data sublayer free for large packets carrying bulky user data. We proposed a SIC receiver structure capable of separating both sublayers with near single-user performance. We demonstrated how to design the operating parameters of such a system based on the SR requirements for reliable detection (that depend on the coding and modulation schemes, as well as channel conditions) and the number of subcarriers. An important result is that time and frequency asynchrony of the users do not seem to have a significant impact on the receiver performance. Furthermore, in the context of WLA the introduced scheme leaves room for a system design backward compatible to existing standards as IEEE 802.11a. The additional signaling component uses the technique to reduce the power per subcarrier, i. e., the interference level it exerts on a simultaneous stream. Thus, using the signaling sublayer will not harm the transmission of legacy devices but brings additional flexibility to more advanced entities. ACKOWLEDGMET This research was supported by the German Ministry of Education and Research within the project Wireless Gigabit with advanced multimedia support (WIGWAM) under grant 01 BU 370. REFERECES [1] J. Heiskala and J. Terry, Wireless LAs: A Theoretical and Practical Guide, Sams Publishing, Indianapolis, 1st edition, 2001. [2] H. Atarashi and M. Sawahashi, Broadband wireless access based on VSF-OFCDM and VSCRF-CDMA and its experiments, in Proc. 03, 4th Int. Workshop on Multi-Carrier Spread-Spectrum, Oberpfaffenhofen, Germany, Sept. 2003, German Aerospace Center, pp. 3 14. [3] S. Simoens, P. Pellati, J. Gosleau, K. Gosse, and C. Ware, The evolution of 5 GHz WLA toward higher throughputs, IEEE Wireless Commun. Mag., vol. 10, no. 6, pp. 6 13, Dec. 2003. [4] K. Claffy, G. Miller, and K. Thompson, The nature of the beast: Recent traffic measurements from an internet backbone, in IET 98, Geneva, 1998, Available at http://www.caida.org/papers. [5] S. Hara and R. Prasad, Overview of multicarrier CDMA, IEEE Commun. Mag., vol. 35, no. 12, pp. 126 133, Dec. 1997. [6] EP BRA, Criteria for comparison, Tech. Rep. 30701F, ETSI, May 1998.