120 1 Performance Analysis of 802.11n Wireless LAN Physical Layer Amr M. Otefa, Namat M. ElBoghdadly, and Essam A. Sourour Abstract In the last few years, we have seen an explosive growth of wireless LAN deployments in both the business and consumer environments. The latest WLAN standard, currently in draft version, is labeled 802.11n. It is based on Multiple-Input Multiple-Output (MIMO) antenna technology combined with Orthogonal Frequency Division Multiplexing (OFDM). MIMO is concerned with spatial diversity and Space-Time Block Coding techniques, which distribute the data signal over several transmit and receive antennas to increase the raw PHY layer data rates to near 600 Mbps, and/or significantly enhance performance in terms of Bit Error Rate (BER). Hence we have set our goals in this research to simulate an 802.11n system in a fading channel, compare the performance of different MIMO modes and channel bandwidths, and study replacing the standard convolutional encoder. T Index Terms Wireless LAN, MIMO Systems, 802.11n, STBC I. INTRODUCTION HE latest improvements to the 802.11 family of standards are achieved using Multiple-Input Multiple-Output Orthogonal Frequency Division Multiplexing (MIMO- OFDM), Space-Time Block Coding (STBC), and by other features such as using double channel bandwidths for transmission [1]. Hence, we realized the need to study in detail the performance of this standard in terms of physical layer (PHY) Bit Error Rates (coded and uncoded) and coded Frame Error Rates versus bit Signal-To-Noise Ratios (SNR), and compare the performance of different MIMO and single antenna modes. Some additional modifications were discussed as well; such as doubling the channel bandwidth from 20 MHz to 40 MHz, and using a convolutional encoder from the cdma2000 standard [2], with a constraint length equal to 9 instead of the 11n standard encoder used with constraint length of 7. Manuscript received October 15, 2007. This work was supported in part by the Faculty of Engineering, Alexandria University. A. Otefa is a Master of Science student in Electrical Engineering in the Faculty of Engineering, Alexandria University, Egypt (phone: +20105205288; e-mail: amr_otefa@yahoo.com). N. ElBoghdadly is with the Electrical Engineering Department, Faculty of Engineering, Alexandria University, Egypt (e-mail: namat@alex.edu.eg). E. Sourour is with the Electrical Engineering Department, Faculty of Engineering, Alexandria University, Egypt (e-mail: sourour@ieee.org). A. MIMO II. BACKGROUND MIMO systems are wireless communication systems in which the transmitter and receiver are equipped with physically separated multiple antennas. The signals are combined in schemes that enhance the error rates and/or data rates for the link [3]. Signals fade independently due to the physical separation between antennas. Examples of some available multiple antenna techniques [4]: Antenna Subset Selection: One or more antennas with the highest power is selected. Also known as Antenna Selection Diversity. Spatial Multiplexing (SMX): Data rate is increased at short distances by transmitting different information simultaneously on the transmit antennas. This method requires a complex receiver. Space-Time Coding: Two main coding schemes; Space- Time Trellis Coding and Space-Time Block Coding (STBC). STBC is a scheme in which the same information is transmitted simultaneously on different antennas. Orthogonal codes are a special case of STBC that can be detected linearly at the receiver with simple operations. Alamouti proposed an orthogonal block code using two transmitters and one receiver (N T N R = 2 1) [5]. Every two time slots, two symbols are transmitted simultaneously over two antennas. This is known as spatial rate 1. At the first symbol time, s 1 is transmitted from antenna 1, and s 2 from antenna 2. Next, -s * 2 is transmitted from antenna 1, and s * 1 from antenna 2 at the second symbol time. The code has full diversity as the same information is transmitted over independent paths to mitigate fading. Using simple linear combining, we can decode the transmitted signal. Only two complex multiplications and one complex addition per symbol are required for decoding. One or more receiver antennas can be used. Other codes were found for N T N R = 3 1 and 4 1 scenarios [6]. Although full diversity, they only convey three information symbols in four channel uses, and thus incur a rate loss (spatial rate 3/4). For N T 1 scenarios, no general construction of orthogonal STBC is known with full diversity and full spatial rate. Moreover, it can be proved [6] that no full rate (i.e., spatial rate 1) orthogonal STBC exists with N T > 2.
120 2 However, for rates < 1, such codes can be found. Several other non-orthogonal and quasi-orthogonal spacetime block codes exist that require full SMX-type detection to recover the transmitted information, but we must compromise gaining advantages through additional diversity on the channel, which might be small for the third or fourth transmit antenna to be added, with this gain at the receiver by making the detection more complicated, eliminating some of the benefits of the additional diversity. The extension to more than two transmit antennas is done by simply filling the additional antennas with zeros (can be viewed as antenna subset selection), repetition, or combination with spatial multiplexing (hybrid STBC/SMX) B. 802.11n The latest WLAN standard, 802.11n, was in draft version the time this paper was authored. Some new features include: STBC and SMX in several configurations: Applicable only when N STS is greater than N SS, N SS spatial streams are mapped to N STS space time streams, which are mapped to N TX transmit chains. These rates are based either on STBC or hybrid STBC/SMX schemes. Denote the complex modulator symbol transmitted on data subcarrier k of OFDM symbols 2n and 2n +1 in spatial stream l as d k,l,2n and d k,l,2n+1 respectively. Table 1 indicates for combinations of N STS and N SS which modulator symbol shall be transmitted during OFDM symbol period (2n) and (2n+1) from space time stream i STS. Provisions for double-width 40 MHz channels for extra throughput. Use of 1-4 spatial streams supporting data rates of up to 600 Mbps. Use of Low-Density Parity Check Codes (LDPCC). Use of a 400ns short guard interval (GI) instead of 800ns GI. III. PERFORMANCE ANALYSIS OF 802.11N WIRELESS LAN PHYSICAL LAYER A. Simulation Model The goal of this research was to study bit and frame error rate performance of PHY layer in the draft 802.11n standard. To do so, a MATLAB model was built to simulate the transmitter, receiver and fading channel in baseband. The transmitter implementation block diagram is shown in fig. 1. Each block represents a MATLAB function. Raw data bits enter the scrambler to prevent long sequences of zeros and ones. In case the PHY rate is greater than 300 Mbps, two FEC encoders may be used. The encoder parser demultiplexes the scrambled bits among the encoders in a round robin fashion. FEC encoders used to encode scrambled bits may be rate ½ convolutional encoders or LDPC encoders. The convolutional encoder used has generating polynomials of 133 8 and 171 8, with puncturing to modify rate to 2/3, 3/4 or 5/6. The stream mapper divides the outputs of the encoders into blocks that will be sent to different spatial streams. We have not simulated rates greater than 300 Mbps to allow linear receiver processing. More complex non-linear receiver designs were out of scope of our simulation. Bits of each spatial stream are interleaved (changed in order) to prevent burst errors, then modulated to complex symbols using BPSK, QPSK, 16-QAM or 64-QAM. Next, the Alamouti STBC scheme is applied on input symbols, pilots are inserted then IFFT is performed to convert to time domain and construct OFDM symbols. Finally a guard interval is inserted between OFDM symbols. In case antenna selection is used, it is done before IFFT. The fading channel was simulated based on the IEEE exponential channel model for indoor multipath. FIR channel taps are returned, given the delay spread and sampling rate. A block diagram of receiver implementation is shown in fig. 2. One or two receive antennas is possible. Guard interval is removed in each antenna, FFT is performed, then pilots are removed. Equalization is performed on the received symbols according to the number of transmit and receive antennas. For systems with single transmit antennas, equalization is done by multiplication by the complex conjugate of frequency domain gains. For multiple transmit antennas, linear processing to decode the Alamouti symbols is performed. Resultant symbols are then demapped to soft values, deinterleaved, fed into the Viterbi decoder and finally descrambled. Input parameters are Modulation and Coding Scheme (MCS), operating channel bandwidth, convolutional encoder constraint length, (7 or 9), length of frame, number of frames simulated, range of SNR simulated, delay spread and mode of MIMO used, whether single antenna systems or MIMO systems. We have varied a set of parameters to realize their effect on PHY layer performance; MIMO mode, channel bandwidth, and encoder constraint length. Fixed parameters were delay spread; set to 100 ns, and frame length; set to 1000 bytes. B. Effect of MIMO Antenna Configurations Several configurations were simulated and compared. These configurations are: 1. Single-Input Single-Output System: This is a typical system with one transmitter and one receiver (1 1). It is a 802.11n standard mode. 2. Single-Input Single-Output System with Antenna Selection: A modification to the above system, adding the support of antenna subset selection. A single antenna is chosen from an array of four transmit antennas according to the sum of channel gains. This may be implemented by channel feedback. The channel with the largest sum of squares of time domain gains is the one whose antenna is selected. This mode is not part of the standard, but was tested to verify if performance gain is worthwhile. 3. Multiple-Input Multiple-Output (Alamouti 2 1): This is the first MIMO mode available in 802.11n. It is an implementation of Alamouti's space-time block coding
120 3 scheme, sending two consecutive OFDM symbols, s 0 and s 1, on transmit antenna number 1, while at the same time s * 1 and s * 0 are consecutively on antenna number 2. This is a standard mode. Receiver processing is linear. 4. Multiple-Input Multiple-Output (Alamouti 2 2): The second MIMO mode available, using two receive antennas instead of one. 5. Multiple-Input Multiple-Output (Alamouti 2 1 with Antenna Subset Selection): Similar to the third mode, but applying antenna subset selection to select two antennas from an array of four or six transmit antennas. This mode is not part of the standard, but was tested to verify if performance gain is worthwhile. 6. Multiple-Input Multiple-Output (Alamouti 2 2 with Antenna Subset Selection): Similar to the fourth mode, but applying antenna subset selection to select two antennas from an array of four or six transmit antennas. This mode is also not part of the standard. A set of simulations were run to compare the performance of different configurations with different Modulation and Coding Schemes (MCS). MCS is a short notation for the used modulation type (BPSK, QPSK, 16-QAM or 64-QAM) and coding rate (1/2, 2/3, 3/4 or 5/6). The first simulation (MCS=7, R=5/6, 64-QAM) compares modes 1, 3 and 4 (1 1, 2 1 and 2 2). Realizations that may be deduced from the plots in fig. 3: For coded BER, gain improvements between 2 1 and 1 1 modes range from 2-4 db, and from 4-8 db between 2 2 and 1 1 modes. For coded FER, gain improvements of 5 db for 2 1 against 1 1 and 10 db for 2 2 against 1 1 at an FER of 10-1, which is the minimum required operating point for WLAN. At lower FERs we have greater improvements. The second simulation (MCS=0, R=1/2, BPSK) compares modes 1 and 2; i.e. the effect of antenna subset selection in single-input single-output systems. Realizations that may be deduced from plots in fig. 4: Improvements at FERs of 10-1 were around 3.5 db. At values below 10-1, the two curves converged indicating a much smaller improvement from antenna selection. The third simulation (MCS=5, R=2/3, 64-QAM) compares modes 3 and 5; thus comparing the effect of antenna subset selection combined with Alamouti STBC (2 1, 4 1 and 6 1). Realizations that may be deduced from plots in fig. 5: For coded BER, an improvement of 1.5-2 db is noticed between 2 1 and 4 1 systems, and 2-3 db between 2 1 and 6 1 systems. Improvements from using 4 1 setups compared to 2 1 setups at FERs of 10-1 were around 1.5 db. Improvements from using 6 1 setups compared to 4 1 setups at FERs of 10-1 were around 0.75 db which is a very small improvement compared to the complexity of adding two more transmit antennas to the standard of four. Similar results to the latter simulation were also found comparing 2 2, 4 2 and 6 2 modes. Based on all the previous simulations we may come to some conclusions regarding the effects of MIMO on wireless LAN systems: MIMO 2 1 antenna arrays yield great performance improvements over 1 1 systems. MIMO 2 2 arrays produce even greater improvements compared to both 1 1 and 2 1 systems. This is due to the fact that two copies of each single symbol transmitted are received. Therefore both full rate and full diversity are achieved when using the Alamouti scheme [5]. Antenna subset selection of 1 or 2 from 4 transmit antennas moderately enhances performance of single antenna systems and MIMO 2 1 and 2 2 systems, whereas if using 6 antennas the improvement is negligible and does not justify the extra complexity for adding two more antennas to the standard of four. C. Effect of Convolutional Encoder Constraint Length The standard convolutional encoder used in 802.11n, the same as in 802.11a, has generator polynomials of g 0 = 133 8 and g 1 = 171 8 and constraint length of 7. It is known that increasing the constraint length enhances the performance of the encoder, but at the same time increases decoder complexity, thus we have a tradeoff between these two parameters; performance and complexity. A convolutional encoder with a constraint length of 9 is used in the cdma2000 standard. This encoder was tested in our WLAN simulation to replace the standard encoder and observe the effects of increasing the constraint length. Its generator polynomials are g 0 = 753 8 and g 1 = 561 8. Realizations that can be deduced from simulation results (MCS=2, R=3/4, QPSK) in fig. 6: For coded BERs, improvements ranged between 0.5-2 db in favor of K=9. For frame FERs, improvements were negligible). Therefore, we may conclude that it is not worthwhile to use an encoder with a constraint length greater than the standard of 7, as we barely have any improvement in FER. The increase in decoder complexity is not justified by any performance gain. D. Effect of Duplicate Operating Channel Bandwidth Channel bandwidth in 802.11n is either 20 MHz (single channel) or 40 MHz (double channel). This feature was added to boost data rates, not enhance performance. We have however studied its effect on error rate performance in one of the simulations (MCS=4, R=3/4, 16-QAM) to verify if it is affected. We realized the following from fig. 7: For coded BERs, a gain of 0.5 db is noticed in low BERs only, in favor of 40 MHz double channels. For FERs, a gain of 0.5 db is noticed in FERs of 10-1 and 10-2 in favor of 40 MHz double channels. Hence, we may conclude that doubling the operating channel bandwidth, although being meant to improve data rates and not error rates, slightly enhances BER and FER for 802.11n
120 4 WLANs. The reason behind this is the increase in bandwidth. Since for frequency selective channels some subcarriers are more faded than others, a greater transmission bandwidth gives a larger variation in gain values. This means a better performance on the average for all subcarriers. However, when the bandwidth is lower, the whole band may fade causing more bit and frame errors. IV. CONCLUSIONS Analysis of all results has brought us to believe that MIMO configurations, specially 2 1 and 2 2 setups, possibly combined with antenna selection from an array of 4 transmit antennas, provided the greatest performance improvement and therefore were our favorable choice. It was also found that it was not worthwhile to use an encoder with constraint length larger than the standard length of 7, as we barely noticed any performance improvement to justify the increase in decoder complexity. We realized that doubling the operating channel bandwidth, despite being meant to improve data rates and not error rates, slightly enhances error rates. Our suggestions for future research may include studying the use of low density parity check codes and turbo codes, hybrid STBC/SMX techniques with more complex receivers, and the correlation between antennas in MIMO systems. REFERENCES [1] IEEE P802.11n /D1.0, Draft Amendment to STANDARD: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications: Enhancements for Higher Throughput, 2007. [2] 3GPP2 C.S0002-C, Version 2, Physical Layer Standard for cdma2000 Spread Spectrum Systems, Revision C, July 2004. [3] D. Gesbert et al., From Theory to Practice: An Overview of MIMO Space-Time Coded Wireless Systems, IEEE Journal on Selected Areas in Communications, Vol. 21, No. 3, April 2003. [4] S. Ten Brink, Coding Over Space and Time for Wireless Systems, IEEE Wireless Communications, August 2006. [5] S. M. Alamouti, A Simple Transmit Diversity Technique for Wireless Communications, IEEE JSAC, vol. 16, no. 8, Oct. 1998, pp.1451-58. [6] V. Tarokh, H. Jafarkani and A. R. Calderbank, Space-Time Block Codes from Orthogonal Designs, IEEE Trans. Info. Theory, vol. 45, July 1999, pp. 1456 67.
120 5 TABLE I STBC MODES IN 802.11N Fig. 1. 802.11n transmitter implementation block diagram Fig. 2. 802.11n receiver implementation block diagram
120 6 (a) (b) Fig. 3. (a) Coded Bit Error Rate (b) Coded Frame Error Rate for 1x1, 2x1and 2x2 systems (MCS=7, R=5/6, 64-QAM)
120 7 (a) (b) Fig. 4. (a) Coded Bit Error Rate (b) Coded Frame Error Rate for 1x1 systems with and without antenna selection (MCS=0, R=1/2, BPSK)
120 8 (a) (b) Fig. 5. (a) Coded Bit Error Rate (b) Coded Frame Error Rate for a 2x1 system with and without antenna selection from 4 or 6 transmit antennas (MCS=5, R=2/3, 64-QAM)
120 9 (a) (b) Fig. 6. (a) Coded Bit Error Rate (b) Coded Frame Error Rate for different constraint lengths; K=7 and K=9 (MCS=2, R=3/4, QPSK)
120 10 (a) (b) Fig. 7. (a) Coded Bit Error Rate (b) Coded Frame Error Rate for different channel bandwidths; 20 and 40 MHz (MCS=4, R=3/4, 16-QAM)