copyright 2002 On the Throughput of Bluetooth Data Transmissions Matthew C. Valenti Assistant Professor Lane Dept. of Comp. Sci. & Elect. Eng. West Virginia University Morgantown, WV Max Robert and Jeffrey H. Reed Mobile & Portable Radio Research Group Virginia Tech Blacksburgh, VA This work was supported by the Office of Naval Research under grant N

© 2002 Motivation & Goals Motivation  Why Bluetooth?  Custom error control = better QoS  Suitable benchmarks? QoS analysis of bottom two layers: Radio/Baseband Goal of this study  To obtain analytical expressions for the maximum average throughput of Bluetooth data transmissions  Can be used as a “benchmark” to compare custom error control protocols and practical receiver designs

© 2002 Overview of Talk Overview of the analysis  Baseband layer analysis Exact maximum average throughput over a binary symmetric channel (BSC) Benchmark for custom error control  Radio layer analysis Upper bound on throughput of noncoherent reception over AWGN and quasi-static Rayleigh fading channels Benchmark for receiver implementation Extensions of this work  Custom error control in Bluetooth  Analysis of other QoS parameters: delay & delay jitter

© 2002 Bluetooth Protocol Stack Application Presentation Session Transport Network Data Link Physical Applications RFCOMM/ Service Discovery Protocol (SDP) Logical Link Control & Adaptation Protocol (L2CAP) Host Controller Interface (HCI) Link Manager Link Controller Baseband Radio OSI Reference Model Bluetooth

© 2002 Bluetooth Baseband Layer Physical links  Asynchronous Connection-Less (ACL)  Synchronous Connection Oriented (SCO) Automatic-Repeat Request (ARQ)  ACL packets use a CRC code and ARQ. Stop-and-wait protocol (single-bit-flag).  SCO packets do not use a CRC or ARQ. Forward Error Correction (FEC)  Packets may be coded with a FEC  Both ACL & SCO can use (15,10) Hamming code.  SCO packets can use a triple repetition code.

© 2002 Frequency Hopping Transmissions are broken into 625  sec slots  Each piconet is synchronized to the master’s clock  Time-Division Duplexing (TDD) Master may only begin transmitting on even indexed slots Slaves may only begin transmitting on odd indexed slots  A transmission may last for 1, 3, or 5 slots Frequency Hopping  Radio hops after each transmission Does not hop during a multislot transmission  Hops through 79 channels Each channel is 1 MHz wide Some countries only allow 23 channels

© 2002 ACL Packet Structure Access Code Payload Packet Header 72 bits54 bits Payload Header Payload Data CRC bits 8 or 16 bits 16 bits bits

ACL Packet Types There are 6 ACL packets that use ARQ: Packet Type Duration (slots) Payload data length (Bytes) Hamming FEC code? Peak throughput DM1117Yes108.8 kbps DH1127No172.8 kbps DM33121Yes387.2 kbps DH33183No585.6 kbps DM55224Yes477.9 kbps DH55339No723.2 kbps

© 2002 Probability of Retransmission A retransmission will occur unless all of the following events occur:  S f : Destination radio synchronizes with the access code of forward packet  H f : Destination radio decodes forward packet header  L f : Destination radio decodes payload of fwd packet  S r : Source radio synchronizes with the access code of the return packet  H r : Source radio decodes the return packet header Thus the probability of retransmission is: P r (  ) = 1 – P[S f ]P[H f ]P[L f ]P[S r ]P[H r ]  where  is the error probability of the demodulator (BER)

© 2002 Synchronization In practice, synchronization is achieved by:  Comparing the hard outputs of the demodulator with a stored copy of the 72 bit access code  Synchronize if the received and stored copies of the access code agree in T bit positions  T is a threshold set to an acceptable false alarm rate. Minimum Hamming distance between distinct access codes is 13, and therefore up to 6 errors can be tolerated. Thus for ML decoding, set T = 66. Therefore synchronization occurs if there are no more than 72-T bit errors:

© 2002 Decoding the Header Header is protected by (3,1) repetition code.  18 information bits (including an 8 bit CRC)  18 three bit code words (triplets) = 54 code bits  Code can correct a single error per triplet Thus probability of correctly decoded header:

© 2002 Decoding the Payload High rate DH packets are uncoded and therefore require that all m bits are correct:  where m=240, 1496, or 2744 Medium rate DM packets have (15,10) single- error correcting Hamming code, thus:  where M=16,100, or 183 is the number of Hamming code words in the payload

© 2002 Number of Transmissions Let N be the total number of transmissions before success  In order to transmit exactly N times First N-1 transmissions must fail Last transmission must succeed  N is a geometric random variable w/ pmf  Where it is assumed that:  is the same throughout a frame (quasi-static) Channel is uncorrelated from Tx to Tx.

© 2002 Average Throughput Data rate is function of N:  D is number of occupied slots (including return) D = 2 for Dx1, 4 for Dx3, or 6 for Dx5 We assume only 1 return slot (asymmetric traffic)  K is number of data bits K= 136, 216, 968, 1464, 1792, or 2712  Average throughput is R avg = E N [R]

Throughput in BSC Channel  Data Rate in kbps DH5 DH3 DH1 DM5 DM3 DM1

© 2002 Bluetooth Radio Layer Bluetooth uses GFSK modulation  Gaussian pulse shaping BT = 0.5  Nonorthogonal frequency shift keying 0.28  h  0.35 We assume h = 0.32  1 Megabaud channel symbol rate  1 MHz occupied bandwidth Reception is normally noncoherent

© 2002 Demodulator Error Rate Because of the ISI induced by Gaussian pulse shaping, an exact expression for noncoherent detection is complicated.  Highly implementation dependent Instead, we use an expression for the exact error performance of noncoherently detected FSK over full- response channels:  This is a lower bound for all receiver architectures

 =E s /N o in dB Data Rate in kbps DH5 DH3 DH1 DM5 DM3 DM1 Throughput in AWGN

© 2002 Quasi-Static Fading Quasi-static fading  SNR remains same for entire transmission.  SNR changes from transmission to transmission. Envelope may be Rayleigh, Rician, or Nakagami  Good model for Bluetooth Short transmissions, uncorrelated channels Retransmission probability pdf of the SNR for Rayleigh fading

 =E s /N o in dB Data Rate in kbps DH5 DH3 DH1 DM5 DM3 DM1 Throughput in Quasi-Static Rayleigh Fading

© 2002 Custom Error Control There is a seventh ACL packet type  AUX1  Occupies one slot  CRC & ARQ are turned off  29 bytes of payload data Can use AUX1 to transport a custom code  Because ARQ shut off, data is delivered from Bluetooth device to host regardless of errors  Perform FEC encoding & decoding on host computer  Implement ARQ on the host computer  No modification of Bluetooth standard is needed

Data Rate in kbps E s /N o in dB BCH coding bound BCH t=10 DM1 DM 3 Example: BCH Coding in AWGN

Adaptive Coding for Quasi-Static Fading E s /N o in dB Data Rate in kbps DM1 DM3 DM5 DH5 BCH10 Adaptive BCH Fully Adaptive

© 2002 Application of Turbo Codes to Bluetooth Turbo codes are capable of achieving near capacity performance. Technical challenges:  Long block lengths needed  Soft-decision decoding is desirable Solutions:  Rate Compatible Turbo Codes (RCPT) Start with a rate ⅓ turbo code 1 frame of data = 4 Bluetooth AUX1 packets There are 8 packets of parity information Only transmit parity as needed  “Pseudo” soft-decision decoding Use the Received Signal Strength Indicator (RSSI)

throughput (kbps) E s /N o in dB TC4 w/ RSSI TC4 no RSSI DM1 DM3 DM5 Performance of Turbo Coded Bluetooth in AWGN

Other QoS Parameters Delay  If packet is transmitted N times, then the delay is  =(DN)(625 x 10-6) Average delay is mean of  Delay jitter is standard deviation of 

E s /N o in dB Avg. delay in msec TC4 w/ RSSI TC4 no RSSI DM1 DM3 DM5

E s /N o in dB Delay jitter in msec TC4 w/ RSSI TC4 no RSSI DM1 DM3 DM5

© 2002 Conclusion Throughput of Bluetooth can be obtained analytically  Exact at the baseband layer  Upper bound at the radio layer Analysis can be extended to other QoS parameters  Average latency & latency jitter  Residual error rate (after timeout) Provides a benchmark  Compare custom error control codes  Compare receiver implementations Further work  More exact expression for BER of demodulator  Implement custom coding using AUX1 packet