Implementation of FEC and MIMO on Wireless Open Access Research Platform Mini Project Final Report
Rupesh Gupta (2006EE50413)
Supervisor Dr. Vinay Joseph Ribeiro
Department of Computer Science & Engineering Indian Institute of Technology Delhi May 2010
Abstract This project was done as a part of the bigger project Rapidly Deployable WiMAX Mesh where the objective is to implement rapidly deployable and self con gurable mesh networks. Such networks are targeted towards disaster management and defence applications. Owing to the vitality of communication in such situations it is highly desirable to ensure robustness of communication to the dynamic environments and channel fades. This project dealt with the physical layer aspects of the WiMAX Mesh. We have used the Wireless Open Access Research Platform (WARP) hardware for all our design implementations. WARP is an FPGA based software radio which provides a scalable and con gurable platform for prototyping wireless communication algorithms developed by the Rice University. Its programmability and exibility makes it easy to implement various physical and network layer protocols and standards. In this report, implementation and performance evaluation of Forward Error Correction (FEC) and Multiple Input Multiple Output (MIMO) antenna mode on the WARP is discussed.
Contents 1 Introduction
2
2 WARP Hardware 2.1 Physical Layer in Hardware . . . . . . . . . . . . . . . . . . . . . . . . . .
4 4
3 Multicarrier Concept 3.1 OFDM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7 7
4 WARP OFDM Reference Design 4.1 Physical Layer Details . . . . . . 4.1.1 OFDM Parameters . . . . 4.1.2 Antenna Con gurations . 4.2 MAC Layer Details . . . . . . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
10 10 10 11 12
5 Error Control 14 5.1 HARQ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 6 Implementation of Type I HARQ on WARP 16 6.1 Physical Layer Design Flows . . . . . . . . . . . . . . . . . . . . . . . . . . 16 6.2 Incorporation of FEC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 7 MIMO 20 7.1 Alamouti Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 8 MIMO on WARP
23
9 Results 24 9.1 Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 10 Future Work
27
1
Chapter 1 Introduction We have observed a maximum range of about 10 meters with the WARP OFDM reference design in lab environment. This range was observed in the Single Input Single Output (SISO) mode keeping the antennas in Line of Sight (LOS) with no obstruction in the path i.e. near perfect channel conditions. The throughput plummets to very low values (20%) with an increase in distance between the communicating nodes beyond this maximum range. Also, the throughput was observed to be highly sensitive to position of the antennas and obstructions in the path. This low range of the WARP boards and sensitivity to antenna placement is partly due to the fact that the reference design does not incorporate any error control mechanism and the use of SISO antenna mode which cannot take advantage of the spatial diversity which may be exploited with the 2 antennas which WARP supports. Here, we have incorporated a convolutional coded Forward Error Correction (FEC) scheme at the WARP physical layer and made use of the mulitple antennas in the Alamouti mode with an aim to increase the range of communication and impart robustness to the varying channel conditions. The performance of each one of these is evaluated against the uncoded SISO reference design and presented at the end for discussion. The remainder of this report is organized as follows. The report is broadly divided into two main sections, the rst one focussing on design and implementation of Type I HARQ at the physical layer and the second one discussing achieving spatial diversity with Alamouti antenna mode on WARP. Chapter 2 details the most important components of the WARP hardware relevant to the scope of this work. In Chapter 3 we give a brief overview of the multicarrier concept with emphasis on Orthogonal Frequency Division Multiplexing (OFDM), the multiplexing scheme used in the WARP reference deisgn. We discuss the important physical and MAC 2
layer details of the OFDM reference design in Chapter 4 which will form the basis for all our further discussions. Chapter 5 describes the basic error control mechanisms viz. FEC and ARQ and a combination of these 2 schemes i.e. HARQ which we have implemented on the WARP. In Chapter 6, the design steps involved in the the implementation of Type I HARQ on WARP are described. The next section discusses the fundamentals of MIMO and space-time codes. Their implementation on WARP is dicussed in Chapter 8 We present our results and discussions in Chapter 9. We end with a brief writeup about the future work in Chapter 10.
3
Chapter 2 WARP Hardware The most important components of the WARP hardware [1] are shown in Figure 2.1 and discussed here brie y • FPGA: An extremely exible and con gurable logic resource. It is used for all types
of processing in the WARP. • FPGA-Radio Interface: This is a present on the radio board and provides A/D
convertors and D/A convertors • Single-Chip Radio IC: This IC is a present on the radio board. It supports dual-band
operation with 2.4 GHz and 5 GHz as center frequencies and a 40 MHz bandwidth independent of carrier frequency. It also performs direct conversion between RF and baseband frequency.
2.1
Physical Layer in Hardware
The PHY layer in hardware [2] is illustrated in Figure 2.2 in which the blue region (Wireless Transmitter and Receiver) represents user designs and red indicates platform support packages. The important support packages in the design are • Radio Bridge: Ties user designs to radio hardware. User needs to instantiate one
bridge per radio board. It ties the ports for user signals (ADC, DAC, gains) and the ports for radio controller I/O to the physical radio ports. • Packet Detector: Detects packets based only on the Received Signal Strength In-
dicator (RSSI) value. The RSSI threshold for accepting a packet can be adjusted from user C code depending on the channel condition. 4
Figure 2.1: Main components of WARP hardware • Automatic Gain Control (AGC): Adjusts the received amplitude by eectively re-
ducing the amplitude if the signal is strong and raising it when it is weaker. The average output of the detected signal level is fed back to adjust the gain to an appropriate level for a range of input signal levels.
5
Figure 2.2: WARP PHY layer in hardware
6
Chapter 3 Multicarrier Concept A digital transmission scheme with linear carrier modulation (e.g. M-PSK or M-QAM) and a symbol duration T occupies a bandwidth B which is typically of the order of T inverse. For a transmission channel with a delay spread S, a reception free of intersymbol interference (ISI) is only possible if S T. As a consequence, the possible bit rate (R proportional to T inverse) for a given single carrier modulation scheme is limited by the delay spread of the channel. The simple idea of multicarrier transmission to overcome this limitation is to split the data stream into K substreams of lower data rate and to transmit these data substreams on adjacent subcarriers, as depicted in Figure 3.1 [3]. This can be regarded as a transmission parallel in the frequency domain, and it does not aect the total bandwidth that is needed. Each subcarrier has a bandwidth B/K, while the symbol duration T is increased by a factor of K, which allows for a K times higher data rate for a given delay spread. 3.1
OFDM
The basic idea of OFDM is to divide the available spectrum into several subchannels (subcarriers). By making all subchannels narrowband, they experience almost at fading, which makes equalization very simple. To obtain a high spectral e ciency the frequency response of the subchannels are overlapping and orthogonal. This orthogonality can be completely maintained, even though the signal passes through a time-dispersive channel, by introducing a cyclic pre x [4]. As shown in Figure 3.2 a cyclic pre x is a copy of the last part of the OFDM symbol which is prepended to the transmitted symbol. This makes the transmitted signal periodic, which plays a decisive role in avoiding intersymbol and intercarrier interference. A schematic diagram of a baseband OFDM system is shown in Figure 3.3
7
Figure 3.1: The multicarrier concept
Figure 3.2: Cyclic Pre x in OFDM
Figure 3.3: OFDM System. CP and CP cross denote the insertion and deletion of the cyclic pre x 8
Chapter 4 WARP OFDM Reference Design The WARP OFDM Reference Design implements a real-time network stack on a WARP node. The design includes a MIMO OFDM physical layer and exible MAC interface for building custom protocols. All processing (hardware control, signal processing, MAC protocol) is executed in real-time by each WARP node. PCs can be attached to WARP nodes via Ethernet for tra c generation/analysis. The design consists of the components [5] shown in Figure 4.1. The main components are as follows • MAC Application: Top-level C code implementing a wireless MAC protocol. • WARPMAC Framework: Low-level PHY control and MAC primitives, implemented
in C code • OFDM PHY: FPGA implementation of the OFDM physical layer, built in System
Generator • Support Peripherals: Other peripheral cores in the FPGA (timer, radio bridges,
etc.)
4.1
Physical Layer Details
4.1.1 OFDM Parameters The WARP physical layer implements a real-time OFDM transceiver in the fabric of a Xilinx FPGA. The OFDM core uses 64 subcarriers, spaced evenly in the transmitted waveform. 4 subcarriers are dedicated to pilot tones [6]. The subcarriers can be loaded with data modulated as BPSK, QPSK, 16-QAM or 64-QAM symbols. Any combination of modulation schemes can be used across subcarriers, though the subcarrier-modulation 9
Figure 4.1: WARP OFDM reference design structure assignment must be xed for a full packet. The reference design loads data into 48 subcarriers. Table 4.1 below lists the number of bytes per OFDM symbol for the supported modulation schemes when using 48 data-bearing subcarriers per symbol. Modulaiton BPSK QPSK 16-QAM 64-QAM
Bytes per OFDM Symbol 6 12 24 36
Table 4.1: Number of bytes per OFDM symbol for dierent modulation schemes
4.1.2 Antenna Con gurations WARP supports two antennas on its two radio boards. The OFDM PHY supports three primary antenna modes: SISO, Alamouti and 2x2 multiplexing. The antenna mode is con gurable per-packet from user C code. The transmitter and receiver must agree ahead of time on the antenna mode.
10
4.2
MAC Layer Details
The Media Access Control (MAC) layer of CSMA OFDM reference design of WARP works as illustrated in Fig 4.2 [7]. It may be described as follows: • At the transmitter:
If the medium is busy, it enters a backo period and waits for the medium to become idle If the medium is idle, it sends the packet, enters a timeout, and waits for an acknowledgment from the receiver • At the receiver:
If the packet passes checksum and is addressed to the receiver, it sends back an acknowledgment • Retransmission of packet:
If sender does not receive an acknowledgement then if the maximum number of retransmits has not occurred, it enters a backo and tries retransmitting. If the maximum number of retransmits has occurred, it drops the packet. The following functions of warpmac.c are called during this process: • The transmitter senses the medium using:
warpmac_carrierSense(-) It return returns a 1 if the medium is idle (and hence the medium can be contented for) and a 0 if the medium is busy (and hence the node must wait). • To transmit a packet over-the-air via the OFDM physical layer, the MAC calls
warpmac_sendOfdm(txBuer) • The transmitter transmits the packet and enters a timeout state:
warpmac_setTimer(TIMEOUT) • The receiver receives a packet via PHY and on the basis of checksum, either:
int receiveGoodPacket(-) is called if the packet passes checksum and an acknowledgement packet is sent to the transmitter which clears its timeout or int receiveBadPacket(-) in the event of receiving a packet that fails checksum and no acknowledgement is sent to the transmitter.
11
Figure 4.2: A state-diagram representation of the transmit receive algorithm
12
Chapter 5 Error Control There are two categories of techniques for controlling transmission errors: the forward error control (FEC) scheme and the automatic repeat request (ARQ) scheme. In an FEC system, an error correcting code is used. When the receiver detects errors in the received codeword, it attempts to determine the locations of these errors. If the exact locations of the errors are determined then the received vector can be correctly decoded, else the received codeword will be decoded incorrectly and erroneous data delivered to receiver. In an ARQ system, a code with good error detecting capability is used. At the receiver, the syndrome of the codeword is calculated, and if the syndrome is zero, the received codeword is assumed to be error free and accepted by receiver. If syndrome is not zero, the transmitter is asked to retransmit the codeword. Erroneous data can be delivered to receiver only if the receiver fails to detect the presence of errors. ARQ is simple and provides high system reliability; however the throughput falls rapidly with increasing channel error rate. FEC systems have constant throughput irrespective of channel error rate; however it is hard to achieve high system reliability because the decoded message must be delivered to receiver regardless of whether it is correct or incorrect. To achieve high system reliability a long powerful code must be used to detect and correct a large number of error patterns which makes decoding hard to implement. 5.1
HARQ
The Hybrid Automatic Repeat Request (HARQ) is a combination of ARQ and FEC [8]. It consists of on FEC subsystem contained in an ARQ system. The FEC increases system throughput and ARQ increases system reliability by sending a retransmission request if the error pattern cannot be corrected. Since the decoder is designed to correct a small collection of error patterns, it can be simple. There are two kinds of HARQ schemes: Type I and Type II. In type I a linear code is 13
used. If the number of errors in received codeword is within the error correcting capability of the code, the errors are corrected and the decoded message saved. If the error pattern is uncorrectable, the received message is discarded and retransmission requested. The type II HARQ scheme is devised based on the concept that parity check digits for error correction are sent to receiver only when they are needed. Two linear codes are used for this scheme; one is a high rate code designed for error detection only and the other is half rate invertible code designed for simultaneous error correction and error detection. First the high rate code is transmitted; if errors are detected only then the half rate invertible code is transmitted. The decoding complexity for type II HARQ is greater than that of corresponding type I HARQ. The disadvantage of type I HARQ is that the overhead due to extra parity-check digits for error correction must be included in each transmission regardless of channel error rate. However type II HARQ is an adaptive scheme which removes this disadvantage.
14
Chapter 6 Implementation of Type I HARQ on WARP We have implemented a type I HARQ on the WARP. As discussed in the previous section, the basic requirement for implementation of a type I HARQ is to incorporate an FEC in the reference design as shown in Figure 6.1. It follows from the discussion in the Section 4.2 that if the error correcting code is unable to correct the error pattern, the received message will be discarded and hence the transmitter will automatically retransmit after it times out.
Figure 6.1: Block diagram for a communication setup with channel coding
6.1
Physical Layer Design Flows
All real-time PHY layer systems are implemented in the FPGA. The real-time design ow starts at the System Generator layer but also must include the layer above, C/C++ code 15
in the PowerPC to implement a MAC to control the physical layer [9]. Additionally, as discussed in Section 2.1, several peripherals like radio bridge connect the physical layers to the hardware.
Figure 6.2: Physical layer design ow
6.2
Incorporation of FEC
The WARP repository contains a verilog implementation of a convolution code FEC [10]. The convolutional coding in this implementation is exactly as per the 802.11a standard and supports rates of 1/2, 2/3 and 3/4. This was integrated with the reference design in the following way: • The manually designed verilog le was converted into a black box using Black Box
Con guration Wizard of Sysgen and inserted at appropriate positions in the sysgen design as shown in Figure 6.3. • The Sysgen model was exported as a peripheral core using EDK Export Tool. • This peripheral core was inserted into the Xilinx Platform Studio project [11]. • The port connections were made in the system.mhs le which contains port infor-
mation to integrate the convolution coded OFDM core with the design. The performance signi cantly improved with the FEC both in terms of percentage of errors as well as range of communication. The iperf results are tabulated in Chapter 9.
16
Figure 6.3: A schematic showing the location of encoder and decoder in the reference design
17
Chapter 7 MIMO Multiple Input Multiple Output (MIMO) is a technology that uses multiple antennas to transmit and receive radio signals and is well suited for systems that have Non-Lineof-Sight (NLOS) functionality. It oers signi cant increases in data throughput and link range without additional bandwidth or transmit power. It achieves this by higher spectral e ciency (more bits per second per hertz of bandwidth) and link reliability or diversity (reduced fading).
Figure 7.1: Various antenna con gurations MIMO can be used in the following ways: • Spatial Diversity: This scheme utilizes two or more antennas to improve the quality
and reliability of a wireless link. In urban and indoor environments, there is not a clear line-of-sight between transmitter and receiver. Instead the signal is re ected along multiple paths before nally being received. Diversity coding sends multiple 18
copies through multiple transmit antennas, so as to improve the reliability of the data reception. If one of them fails to receive, the others are used for data decoding. • Spatial Multiplexing: This scheme exploits the spatial dimension to increase through-
put without any additional bandwidth. This is done by transmitting independent and separately encoded data signals, so called streams, from each of the multiple transmit antennas. Spatial multiplexing requires statistically independent channels between each antenna pair, which is usually achieved by lots of scattering (and minimal line-of-sight) in the propagation environment.
Figure 7.2: Spatial multiplexing in MIMO
7.1
Alamouti Mode
Alamouti is a space time block code (STBC) which is designed to achieve reliability by exploiting spatial diversity. It is a 2x1 MIMO scheme and has the coding matrix as given in Figure 7.3
Figure 7.3: Alamouti coding matrix It is readily apparent that this is a rate-1 code. It takes two time-slots to transmit two symbols. It is the only orthogonal STBC that achieves rate-1. That is to say that it is the only STBC that can achieve its full diversity gain without needing to sacri ce its data rate. 19
Chapter 8 MIMO on WARP As mentioned in the section 4.1.2, WARP supports two antennas on its two radio boards. The core has ports for two antennas and supports three primary antenna modes: SISO, Alamouti and 2x2 multiplexing. • SISO: This is the simplest mode, wherein the transmitter sends and the receiver
captures all data from a single antenna. The choice of transmit antenna can be set from user code per-packet. The choice of receive antenna can be either set perpacket by code or automatically per-packet by selection diversity logic. In selection diversity mode, the receiver checks the AGC gain choices for both antennas and chooses the stream from the antenna with the higher SNR (i.e. lower AGC gains). • 2x2 Multiplexing: This mode uses spatial multiplexing to transmit and receive from
both antennas. This scheme doubles the data rate (vs. SISO), as two symbols are transmitted with each channel use. But this mode is less reliable than SISO or Alamouti, since a given data bit is transmitted from only one antenna (diversity = 1). • Alamouti: This mode uses Alamouti's simple transmit diversity scheme. Both trans-
mit antennas are used for every packet, with all data being sent from both antennas, encoded with the Alamouti STBC. The receiver uses a single antenna, which can be selected by user code or by selection diversity. The PHY modes (SISO, Alamouti and multiplexing) do not interoperate over-the-air and hence every node must be con gured with the same PHY mode. The antenna mode (choice of SISO antenna, selection diversity, etc.) can be con gured per-node. We have implemented both 2x2 Multiplexing and Alamouti the results of which are tabulated in Chapter 9 . 20
Chapter 9 Results We have used Iperf network testing tool to measure the throughput of communication between 2 WARP nodes. The results are tabulated below for dierent cases and discussed next. • Uncoded CSMA MAC with SISO antenna mode in LOS: Table 9.1 corresponds to a
distance of 2 meters between the 2 nodes in a quiet lab environment. The maximum range (successful ping) that could be achieved in this con guration was around 10 meters when the antennas were placed at the same height. No communication could be observed when the LOS path was obstructed with glass or metal. Rate (Mbps) 1.19 2.38 3.55 4.76 5.95 7.09
Error % .08 .12 .20 .24 .26 .94
Table 9.1: Throughput of uncoded CSMA MAC, SISO, LOS, 2m distance • Coded CSMA MAC with SISO antenna mode in LOS: Table 9.2 corresponds to a
distance of 2 meters between the 2 nodes in a quiet lab environment. The maximum range (successful ping) that could be achived in this con guration was around 10 meters when the antennas were placed at the same height and the LOS path obstructed with glass. As can be observed from Table 9.3, even with a glass barrier we could stream 1 Mbps video with imperceptible error of around 2%. No communication could be observed when the LOS path was obstructed with metal. • Uncoded CSMA MAC with Alamouti antenna mode in NLOS: Table 9.4 corresponds
21
Rate (Mbps) 1.19 2.38 3.55 4.76 5.95 7.09
Error % 0 0 0 .029 .14 .24
Table 9.2: Throughput of coded CSMA MAC, SISO, LOS, 2m distance Rate (Mbps) 1.19 2.38 3.55
Error % 1.9 4.5 6.0
Table 9.3: Throughput of coded CSMA MAC, SISO, Glass obstructed, 10m distance to a distance of 2 meters between the one set of antennas and 3 metres between the other in a quiet lab environment. Rate (Mbps) 1.19 2.38 3.55 4.76 5.95 7.09
Error % 0 1.8 4.2 6.1 10 11
Table 9.4: Throughput of uncoded CSMA MAC, Alamouti, NLOS • Uncoded CSMA MAC with 2x2 multiplexing antenna mode in NLOS: No eective
communication could be achieved between the nodes. The communication was found to be highly sensitive to the placement of antennas. We could successfully ping at only one particular location of the antennas with an error percentage of around 10% at 1 Mbps data rate.
9.1
Discussions
It can be readily observed from the above results that coded physical layer outperforms the uncoded one and allows communication at larger distances and NLOS with minor obstructions in the path. However the Alamouti does not yeild results as expected. Alamouti was expected 22
to achieve reliability (lesser error compared to SISO) by exploiting spatial diversity as mentioned in Section 7.1. Also, 2x2 multiplexing failed to achieve high data rate. This behaviour can be attributed to the fact that spatial multiplexing requires statistically independent channels between each antenna pair, which can be achieved by lots of scattering and minimal LOS in the propagation environment. This is di cult to achieve in a quiet lab environment. Apart from that, the physical layer implementation of multiplexing in the reference design is very simple, using a zero forcing equalizer at the receiver. A zero forcing equalizer multiplies the input signal by the reciprocal of channel transfer function of the particular channel. This is intended to remove the eect of channel from the received signal, in particular the intersymbol interference (ISI). The zero-forcing equalizer is ideal when the channel is noiseless. However, when the channel is noisy, the zero-forcing equalizer will amplify the noise greatly at frequencies where the channel response has a small magnitude. Hence the communication system will be highly sensitive to changes in the propagation environment and antenna placement. This explains the complete failure of 2x2 multiplexing at certain antenna positions.
23
Chapter 10 Future Work The coded physical layer has been successfully tested with Aloha and CSMA MAC layers over SISO link. It still needs to be tested with the TDMA MAC layer developed as another subpart of the project Rapidly Deployable WiMAX Mesh. A more rigorous analysis needs to be done for the failure of 2x2 multiplexing mode and the lower reliability of Alamouti mode compared to SISO. It would be higly desirable to have an FEC implemented on MIMO for high reliability and bandwidth e ciency. This would require a re-design of the full decoder pipeline. The decoder would have to process bits at twice the datarate, which would aect the architecture used in the Verilog physical layer design.
24
Bibliography [1] Patrick Murphy, WARP University, Nov 2008
Hardware and Design Flows,
WARP Workshop, Rice
http://warp.rice.edu/trac/wiki/Workshops/Rice_2008November
[2] Siddharth Gupta, WARP Real-Time PHY Design Flow, WARP Workshop, Rice University, Nov 2008 http://warp.rice.edu/trac/wiki/Workshops/Rice_2008November
[3] Henrik Schulze, Christian Luders, Theory CDMA, John Wiley Sons, Ltd, 2005
and Applications of OFDM and
[4] Edfors, O., Sandell, M., Van de Beek, J.-J., Landstrom, D., and Sjoberg, F., An Introduction to Orthogonal Frequency Division Multiplexing, Lulea, Sweden, Lulea Tekniska Universitet, 1996 [5] WARP OFDM Reference Design http://warp.rice.edu/trac/wiki/OFDMReferenceDesign
[6] OFDM Physical Layer Details Documentation http://warp.rice.edu/trac/wiki/OFDM/MIMO/Docs/PHYDetails
[7] Carrier-Sense Medium Access Reference Design Documentation http://warp.rice.edu/trac/wiki/CSMAMAC
[8] Shu Lin, Daniel J. Costello,
Error Control Coding,
Prentice Hall, May 2004
[9] Physical Layer Design Flows http://warp.rice.edu/trac/wiki/DesignFlowPHY
[10] Convolution Coded Verilog Model http://warp.rice.edu/trac/browser/ResearchApps/PHY/MIMO_OFDM/ ConvCoded
25
[11] Tutorial on Using a Custom Peripheral in XPS http://warp.rice.edu/trac/wiki/Exercises/UsingCustomPeriphs
26
