6. CONCLUSION

1. A NETWORKING PERSPECTIVE OF MOBILE PARALLEL RELAYS IEEE Workshop on DSP, Taos Ski Valley, NM, 1-4 August 2004 YINGBO HUA, YU CHANG, YAN MEI University of California, Riverside, CA, 92521, USA . q We can show that for large SNR, 1. MOTIVATION AND CHALLENGE 5. THROUGHPUT - Con’t 4. PACKET TRANSMISSION - Con’t 5. THROUGHPUT - Con’t • q Tier t transmits a new packet only after an instruction is received from tier t+1. • q The control (scheduling) signals may be transmitted via channel(s) orthogonal to the channel(s) used for data transmission. • qIn mobile ad hoc network, how to achieve and maintain robust connection with constraint on delay, power and bandwidth? • q High mobility of nodes and interfering objects causes fast fluctuations of SNR. • Frequent power control or frequent network routing based on SNR (or equivalent) feedback may not be efficient. where q For , d(t)=N2, i.e., the diversity gain is N2. 5. THROUGHPUT AND POWER SAVING q Note that with only two relays at each link, we have that q Mobile parallel relays (MPR) with space-time modulation/coding: . 2. A DSP SOLUTION: MPR Figure 3: The lower bound of throughout gain from serial relaying to parallel relaying as function of the packet loss rate between two single nodes. • Assume: • q There are N parallel relays in each tier except at the source and the destination. • The channel fading between a transmitting node and a receiving node at each link is statistically independent and identically distributed (i.i.d.) across all transmit-and-receive pairs within the link, and all links in the network have the same statistical property. • The channel fading factor is constant during a single transmission of a packet. But the fading factor may change from packet to packet, including retransmitted packets. (Multiple packets may be transmitted over multiple carriers of different fading factors.) • The bit error rate (BER) at the ith receiving node in response to k transmitting nodes is denoted by pi(k). q To consider the power saving from serial relaying (N=1) to parallel relaying (N>1), we let the total transmission power to be upper bounded by a constant independent of N. The orthogonal space-time modulation is used in all cases. q Figure 4 shows the PLR at tier 5 for different values of N at each tier. We see a significant power saving by using N > 1. q Consider the use of Golay (24,12) code for error correction and detection. The following figure shows the packet loss rate (PLR) as a function of SNR where two relays are used at each tier. The Alamouti code is used for space-time modulation. QPSK symbol modulation is assumed. The top curve in Figure 2 is the PLR at tier 1, with a slope equal to 2. The second curve is the PLR at tier 2, with a slope equal to 3. The third curve is the PLR at tier 3, with a slope equal to 4. All other curves have the slope equal to 4 as predicted by theory. • q A cluster of nodes present in a small neighborhood may serve as a set of MPR for one link of a route. • In transmission mode, space-time modulation or codes may be used by MPR as if by multiple transmitters. • The spatial diversity gained by MPR reduces delay and power consumption. • q Synchronization among MPR is a unique DSP issue currently under investigation by a joint UCR/UCLA team. • . Then: q With a full-diversity space-time modulation method used at the ktransmitting nodes, the averaged BER over Rayleigh fading channels for a class of symbol modulation methods is known to be proportional to for large (averaged) SNR. q If a block code and a hard decision decoding are used, the packet loss rate (PLR) is 3. FINDING A ROUTE OF MPR • Assume that each node knows its next-hop neighbors for a given destination D. • qStep 0: Set the group (tier) index t=0. Since the source knows the next-hop relays for the destination, we can now assume that the group t of parallel relays Rt(1), Rt(2), …, Rt(Nt) all know their next-hop relays , Rt+1(1), Rt+1(2), …, Rt+1(Nt+1). (Note N0 =1) • qStep 1: The lead relay Rt(1) requests Rt+1(1), Rt+1(2), …, Rt+1(Nt+1) to provide (in order) their tables of next-hop neighbors with reference to D. Rt(1) figures out Rt+2(1), Rt+2(2), …, Rt+2(Nt+2) as the intersect of those tables, and then provides this information to Rt+1(1), Rt+1(2), …, Rt+1(Nt+1). • qStep 2: Set t=t+1. Go to Step 1. Figure 4: Packet loss rates at tier 5 for different numbers (N) of parallel relays at each tier. The total transmission power at each tier is kept to be independent of N. Figure 2: Packet loss rates at different tiers where two relays are used at each tier • q To keep the PLR at 1% from N=1 to N=2, there is almost a 10 dB reduction of SNR. • q It is important to note that from N=1 to N=2, there is no additional cost of bandwidth. • Beyond N=2, there is a penalty factor of bandwidth (between one and two) for using orthogonal space-time modulation. • q The power saving becomes insignificant when N >4. • We can show that for large SNR, the averaged PLR is also proportional to • qAssuming that the source has only one transmitter, the average probability that g receiving relays in tier 1 lose a packet is where c is the number of correctable error bits. • q  The average delay at each link can be shown to be • where D is the time of each transmission of a packet, • successfully or not. • q The network throughput can be measured by a normalized inverse of Tt(N), i.e., • Assuming Pk+1< Pk , the gain of the network throughput from serial relaying to parallel relaying is • which is significant whenP1 is not very small. • q Figure 3 illustrates the lower bound of the throughput gain. qMore generally, for the relays in tier t and , the average probability that g receiving relays lose a packet can be shown to be 4. PACKET TRANSMISSION 6. CONCLUSION • q A packet in tier t is repeatedly re-transmitted to tier t+1 until the packet is received correctly by at least one relay in tier t+1. • Any relay in tier t+1 can initiate the transmission (forwarding) of a packet to tier t+2 upon successful reception of the packet from tier t. • When tier t hears a forwarded packet or an ACK from t+1 , tier t decides that tier t+1 has received the packet correctly. • Mobile parallel relays (MPR) are feasible from the networking point of view. • q  The potential benefits of MPR are significant. • q With a reduced burden on networking traffic, MPR may yield more than 10 dB power saving during data transmission in a highly mobile environment. • q MPR also hinges on successful DSP implementations. where the denominator 1-Gt-1(N) is due to the fact that the relays in tier t-1 do not forward a packet until at least one of them received the packet correctly q It is clear that Gt(N) measures the average probability of link failure in tier t. This work is supported in part by the ARL’s CTA program.