A spread spectrum DS/CDMA channel fading simulation environment for mobile radio network systems

全文

(1)models CH changes Cluster switches. distance 0.010 0.033. fading/hys 0.564 1.008. Table 3: Impact on clustering algorithm reservation algorithm each packet in the VC stream is routed individually, based on the destination address, very much like a data packet. As a di erence, however, the

(2) rst packet in the VC stream, upon successfully capturing a slot, will reserve it for all subsequent frames. If the slot remains unused for a certain number of frames, it is declared free by the CH and it is returned to the free slot pool. For VC routing, the distributed Bellman-Ford routing scheme based on shortest paths is used. Similar to what we did in the previous section, we compare the performance using simulation under the channel fading model and the distance model. We use the same simulation boundary and the same mobility model as before. This time we put 20 nodes inside the square. Poisson call arrivals, exponential duration times and constant bit rate trac within the call are assumed. First, we study the impact of channel fading on the clustering algorithm. We monitor the cluster changes when nodes move. Cluster changes are good indicators of the underlying topology stability. Two measures account for cluster changes. One is \CH changes", which is the number of nodes which change the role as CHs. Another is \cluster switches" which is the number of nodes which switch from one cluster to another. We run the clustering algorithm and monitor cluster changes per unit time in both distance and fading channel models. Table 3 reports the changes. We found out that the number of changes observed using the channel fading model is signi

(3) cantly larger than that yielded by the distance model. The \ampli

(4) cation" e ect due to fading (with respect to the distance model) is 30. Thus, in such an indoor environment, a very fast power adjustment would be required to stabilize link quality such that the link is not declared up and down intermittently. Alternatively, some careful averaging technique must be developed to smoothen out the uctuations. Next, we study the impact of channel fading on the ability to dynamically reroute the VC when changes occur, and thus avoid packet loss. We

(5) nd that since the equivalent mobility has been dramatically ampli

(6) ed, the Bellman-Ford routing scheme is not able to catch up with the speed of topology change. The total fraction of lost packets for the fading model is 30 times larger than that for the distance model. models pkts dropped(including duplication) (pkts dropped by no route). distance 0.2% (0.03%). fading/hys 6.16% (2.78%). Table 4: Impact on clustering performance. Up to 50% of the total lost packets are dropped due to lack of up to date routing information at the time when the VC must be rerouted after failure of the current path. Obviously, the distributed shortest path algorithm is not adequate to handle such rapid uctuations. A routing scheme which can withstand rapid changes in link state (ON/OFF) is needed.. 6 Future Work and Conclusions An indoor channel fading model based on SIRCIM has been implemented in a simulation environment. This model accounts for small-scale as well as large-scale time and space correlation of the radio signal. It calculates the received power for each packet, and allows the accurate modeling of propagation e ects in a mobile radio network simulation. We have investigated the impact of channel fading on the design of network algorithms, and have observed severe degradation of performance in moving from free space to channel fading models. The results point to the need of

(7) nding an appropriate power adjustment algorithm which can compensate for the fast fading channel, and a routing algorithm which remains stable in spite of power uctuations due to fading.. References T. S. Rappaport and S. Y. Seidel. 1990. \SIRCIM: Simulation of Indoor Radio Channel Impulse Response Models." VTIP, Inc. T. S. Rappaport, S. Y. Seidel, and K. Takamizawa. 1991. \Statistical Channel Impulse Response Models for Factory and Open Plan Building Radio Communication System Design." IEEE Trans. on Comm., vol.39, no.5 (May):794-807. Joel Short, Rajive Bagrodia, and Leonard Kleinrock. 1995. \Mobile Wireless Network System Simulation." ACM Mobile Computing and Networking Conference. (will appear). Jorgen Bach Andersen, Theodore S. Rappaport, and Susumu Yoshida. 1995. \Propagation Measurements and Models for Wireless Communications Channels." IEEE Communications Magazine, (Jan):42-49. Homayoun Hashemi. 1993. \Impulse Response Modeling of Indoor Radio Propagation Channels." IEEE J. Select Areas Comm., vol.SAC-11, (Sep):967-977. M. Gudmundson. 1991. \Correlation Model For Shadow Fading In Mobil Radio Systems." Electron. Lett., (Nov):2145-2146. Scott Y. Seidel, Koichiro Takamizawa, and Theodore S. Rappaport. 1989. \Application Of Second-Order Statistics For An Indoor Radio Channel Model." IEEE Vehic, Tech. Conf., Proc., (May):888-892. C. Chien et al. 1994. \A 12.7 Mchips/sec All-Digital BPSK DirectSequence Spread Spectrum IF Transceiver." IEEE Journal of Solid-State Circuits, vol.29, no.12, (Dec). Mario Gerla, Jack Tzu-Chieh Tsai, Nicholas Bambos and Shou C. Chen. 1995. \A Distributed, Mobile Wireless Infrastructure for Multimedia Applications." Fifth WINLAB Workshop on Third Generation Wireless Networks, New Jersey, (Apr). Mario Gerla, and Jack Tzu-Chieh Tsai. 1995. \Multicluster Mobile Multimedia Network." ACM-Baltzer Journal of Wireless Networks, (Aug). (will appear). D. J. Goodman and S. X. Wei. 1991. \Eciency of packet reservation multiple access." IEEE Trans. on Vehic. Tech., vol.40, (Feb)..

(8) will declare that the link is in the ON state if the distance is within 250 units. We monitor the state average durations and the transitions out of a state during 20,000 time units. Table 1 reports the results. The number of ON/OFF transitions per time unit is calculated by dividing the total number of state transitions by 20,000 time units. It represents the frequency of link state models ON/OFF transitions per time unit mean duration of a state. distance. fading. 0.00122. 0.06133. 496.52. 16.78. in Table 2, we discover that the e ect of fading is signi

(9) cantly reduced by introducing hysteresis, as expected. models ON/OFF transitions per time unit mean duration of a state. fading. fading/hys. 0.06133. 0.0198. 16.78. 120.25. Table 2: Impact on link level by using hysteresis threshold. 5.2 Case study: WAMIS clustering algorithm. Table 1: Impact on link level change which is also a good indicator of topology change. The mean duration of a state is the mean duration that the link stays in the same state. It is basically an indicator of link stability. It is obvious that the fading ampli

(10) es mobility. In our example, for the given choice of parameters the ampli

(11) cation factor ranges from 30 to 60. This result is con

(12) rmed by Fig. 5, where received power is plotted as a function of time. Next, we investigate the impact of the channel fading model on the performance of the WAMIS clustering algorithm (Gerla et al. 1995; Gerla and Tsai 1995). The WAMIS architecture supports VCs for real time (voice, video) traf

(13) c with guaranteed delay in mobile radio environments. In this scheme, neighboring nodes are grouped into clusters. Each cluster has one cluster head (CH) which is dynamically elected by the cluster members according to the lowest-ID rule. Fig. 6 shows a 10 node example, where nodes 1, 2 and. 65 5. 2. received power (dB). 10. 60 8. 55. 1 9. 6 3. 50 4. 7. 45. 40 50. 100. 150. 200. 250 time. 300. 350. 400. 450. 500. Figure 6: Example of cluster formation (lowest-ID) Figure 5: Power received during the

(14) rst 500 time units for the

(15) rst 500 time units. Note that the mean value of the curve approximately represents the distance model, while the spikes are caused by multipath fading. This is a very dramatic e ect which cannot be ignored in the design of mobile wireless networks. One way to reduce the ampli

(16) cation e ect is to use a hysteresis threshold for link activation. In other words, once a link is switched to OFF, it can be turned to ON again only when the received power is above threshold plus . Once the link is ON, it will stay ON as long as the received power is above threshold. It becomes OFF when the received power is below threshold. In this way, we can avoid using links which can be intermittently heard due to channel fading, but in general are not stable enough. We have not developed yet a general criterion for choosing , but in the following simulation, we choose to be the standard deviation of the received power on a link 250 units long. From results. 4 are CHs. The CH has the role of regional broadcast node, and of local coordinator to enhance channel throughput. The store-and-forward multihop sessions across di erent clusters will be carried through by GWs which are connected with two or more CHs. A TDMA frame is de

(17) ned inside a cluster. Bandwidth is guaranteed to a real time connection by assigning a given number of slots in the TDMA frame. The remaining free slots in each cluster can be accessed by datagram trac using an S-ALOHA scheme. Di erent sets of codes are assigned to neighboring clusters in order to limit interference across clusters. By this clustering architecture, a mobile instant infrastructure of wireless radio stations is implemented which enhances spatial reuse of time slots and codes (Gerla et al. 1995; Gerla and Tsai 1995). In this experiment, we focus on real time trac support. A fast reservation VC scheme described in (Gerla et al. 1995; Gerla and Tsai 1995) and inspired by PRMA (Goodman and Wei 1991) is used to carry voice/video packets. In the fast.

(18) Data. Packet Size. Large Scale. Small Scale. Correlation. Correlation. Gold Cod. Power. Chip Rate. 4.3 SIR measurement. Impulse Response. SIR/Power Measurement. Processing Gain. Figure 3: SIR/Power measurement. 4.1 Packet Generator and PN sequence Each node will be represented by TRANSCEIVER (status, code, power,location) which describes the status (transmitting or receiving), code (PN sequence), power level and location. When a node transmits, it generates one packet which consists of a random string of bits. The bit string is spread by the speci

(19) c spreading code and transmitted to the receivers.. 4.2 Channel Impulse Response Generation For each active link (with small T-R separation), the impulse response must be simulated. Initially, we draw a value of the parameter A1 as the amplitude of the

(20) rst path in the impulse response using the SIRCIM model. Then, we use temporal correlation coecients described in (Rappaport et al. 1991), and apply Eqs. 4 and 5 to generate the subsequent multipaths to form the entire impulse response pro

(21) le (see Fig. 4). The initial positions of transmitter and receiver as. power loss (in dB). 0. -20. -40. -60. After the updated impulse response is generated, we can convolve it with our random-chip sequence to compute the received power. In our model, the chip rate is 12.7 MChips/sec (Chien et al. 1994). The receiver can only resolve the multipath components with excess delay larger than the chip time, (in our case, 79 ns). All components with excess delay less than 79 ns are seen by the receiver as a single component (the main path component). Their sum, accounting for the various phases, determines the received amplitude. A rejection receiver which locks on the strongest path component is assumed. All the other path components will produce intersymbol interference. We can compute the e ective received power at the target receiver for all transmitters (including same code and di erent code). Next, for the target receiver, we compute the SIR (signal to interference ratio) for each packet. Here the interference consists of two contributions: (a) the interference caused by di erent coded packets, and; (b) the intersymbol interference caused by the packet itself (i.e. multipath).. 5 Experimental Results Intuitively, we expect that in a mobile radio environment channel fading has the same e ect as increasing the radio speeds as far as received power is concerned. This is because, even when the mobile moves just a few feet, the channel attenuation changes so drastically that it looks like the mobile had moved quite a large distance. In other words, channel fading \ampli

(22) es" the e ect of movement on the rate of change of received power. How much does channel fading impact mobile radio network performance? To answer this question, we

(23) rst investigate the impact on network connectivity (i.e. topology). Next, we investigate the impact on throughput eciency on an individual Virtual Circuit (VC) in the WAMIS (Wireless Adaptive Mobile Information Systems) TDMA clustering scheme.. 5.1 Connectivity Experiment. -80. -100 0. 50. 100. 150. 200 250 300 excess time. 350. 400. 450. 500. Figure 4: Impulse response pro

(24) le well as the impulse response are memorized and later used to compute signal levels after radio displacement. Namely, when the radios move, we apply the large-scale or the smallscale correlation properties to get the new impulse response from that of the previous positions as a function of the displacement (see Fig. 2). The number and distribution of the multipath components generated by SIRCIM will determine the change of the environment after displacement. Therefore, the new condition of the active link after displacement can be characterized by the updated impulse response.. The mobility boundary is a 500500 square. Distance unit is assumed to be 2 feet, thus the area is 10001000 feet, a typical operational theater for WAMIS applications. We put a transmitter and a receiver inside it. Fixed transmit power is assumed. Between each received power calculations, each node can move (with probability 0.1) 5 units in both xand y- axes with equal likelihood. Thus, we must account for both shadowing and multipath channel fading. The aspect which is likely to have the most impact on protocol design is the change of topology with mobility. Since the topology of a wireless network consists of point to point links, we can simplify the problem by monitoring the stability of a single link instead of that of the entire topology. The link is in the \ON" state if the received power is above a given threshold; otherwise, it is in the \OFF" state. The threshold between ON and OFF is the mean received power when two nodes are 250 units apart. Therefore, the distance model.

(25) tance d0 , and n is a path loss exponent. X is a zero-mean Gaussian distributed random variable with standard deviation . In wireless channel modeling, a common assumption is that shadowing is independent from one location to another. Unfortunately, this assumption is not valid in a dynamic model with mobile users, where location dependent correlation must be accounted for in order to provide continuity. In this paper, we include the correlation model for shadow fading developed in (Gudmudson 1991). We adopt a correlation function which covers a range of up to 10 meters, with correlation coecient typically ranging between from 0.9 to 0.1. When one node moves around, the mean path loss PL(d) will be calculated based on the correlation between the previous site and current site.. 3.2 Multipath Fading. The radio signal received at the receiver is typically the sum of a large of number of rays, each of which is characterized by its own attenuation, delay and phase. In an indoor environment, the signal level hd (t), i.e. channel impulse response, received at time t is given by the aggregation of the signals on the various paths. More speci

(26) cally:. hd (t) =. X (d)e N. n=1. n. ?jn (d) (t ? n ). (2). where N is the number of multipath components, n represents the amplitude of component on the n-th path, n is the phase shift and n is the time delay of the n-th path with respect to the

(27) rst arriving component, d is the distance between transmitter and receiver, and (t) is the Dirac delta function. Ref. (Rappaport et al. 1991) describes wide-band, in-building propagation models for factories and open plan buildings which provide: (1) the distribution of the number of multipath components; (2) the probability of receiving each multipath component; (3) the distribution of the amplitudes, phase, time delays of multipath components. The SIRCIM model assumes that the amplitude of an individual multipath component obeys log-normal statistics. Namely, the amplitude Ak (Tk ; D) of the k-th path with excess delay Tk and distance D can be characterized as 2 Ak (Tk ; D) = N [N [10 n(Tk ) log( 2D:3 ); large ?scale]; 2 small ?scale(Tk )]. (3) 2 where N [x, x ] represents the normal distribution with mean x (in DB) and a standard deviation x2 (in DB), n denotes the power law relationship between distance and received power. The amplitude Ak of k-th path with excess delay Tk will obey both large scale and small scale fading variation.. 3.3 Autocorrelation Coecient Function. A number of impulse response pro

(28) les collected in the same local area are generally similar since the channel characteristics do not change appreciably over short distance. On the. large scale correlation. ( 10 meter ). receiver receiver small scale correlation ( 1.5 meter ). Transmitter. Figure 2: Large/Small-scale correlation other hand, when the receiver moves to other local area as shown in Fig. 2, the impulse response pro

(29) le will change according to the large scale correlation function. For two variables with log-normal distributions, the jointly log-normal distribution property can be applied (Rappaport et al. 1991; Seidel et al. 1989). Namely, given the mean power of a multipath component at one location (or for one excess delay) 1 , we can estimate the mean power of the same component at the other location (or for the other excess delay) 2 using correlation coecients as follows:. A(2 )jA(1 ) = A(2 ) + (1 ; 2 )(A(1 ) ? A(1 )) A(2 ) (4) A(1 ). Likewise, we can estimate the standard deviation of the amplitude as:. A2 (2 )jA(1 ) = (1 ? 2 (1 ; 2 ))A2 (2 ). (5). where A is the amplitude, represents the spatial location (or excess delay), and is the spatial (or temporal) correlation function of multipath amplitudes. Using the above results, we extend the model by integrating the spatial correlation and temporal correlation observed in (Rappaport et al. 1991). Since the mean individual multipath component powers are log-normal distributed also over large scale systems, we can use jointly log-normal distribution to estimate the next impulse response pro

(30) les by using the spatial autocorrelation coecient.. 4 The Approach In the DS/CDMA environment, each data bit is spread (i.e. encoded) into a sequence of chips (e.g. a gold code sequence). To exploit CDMA, di erent transmissions can use nearly orthogonal gold codes. Thus, if two or more packets with di erent codes are received at the same time, the decoder captures the packet with its same code, and interprets the other packets as noise. Therefore, the former contributes to the \signal" power, and the latter contributes to the \interference" power. Signal to interference ratio (SIR) can thus be derived. The entire simulation/measurement procedure is shown in Fig. 3..

(31) a relocation has taken place. This tool computes the received signal power as needed in network protocol simulations. It is integrated in a modular fashion into the simulation environment, thus allowing us to investigate the impact of channel fading on network algorithms. The rest of the paper is organized as follows. Section 2 will describe the simulation environment and the capability of this environment. Section 3 will

(32) rst introduce the channel fading models which are derived from SIRCIM models. The simulation implementation is described in Section 4. The impact on network protocol evaluation is illustrated with an example in Section 5. Finally, future work and conclusions are in Section 6.. 2 Simulation Environment We have developed a general purpose, parallel environment (Short et al. 1995) for the simulation of network algorithms and their evaluation (e.g. stability and performance as a function of the radio characteristics, network control, and mobility pattern). The main thrust of this paper is the wireless channel implementation in the simulator and its impact on overall wireless network performance evaluation. With a realistic channel model, the simulation environment can be used to fully investigate the impact of radio propagation e ects (e.g. multipath fading) and radio mobility on existing network algorithms and protocols.. 2.1 Channel Modeling for Mobile Radios. The channel model is a very important part of the mobile radio network simulation. It is responsible for determining which nodes are able to communicate with each other and what the quality of transmission is. If nodes are moving, the link state will change continuously during the simulation run. For example, the link may fail due to low transmitting power; also, it may experience the e ect of shadowing, and multipath fading. Likewise, the link quality may drastically decay due to obstacles. Thus, a rather sophisticated channel model is required to capture the e ect of mobility and its impact on network algorithms.. Network. Node4 Node3 Node2 Node1 Routing Connection. Wireless Channel. Link Quality. Modeling Power Control. Packet Driver Mobility. location. Figure 1: Network simulation for in the proposed simulation environment as shown in Fig. 1. The channel fading models and the way to predict the link quality will be described in the next two sections.. 3 Channel Propagation Models Radio channel propagation is characterized by three main components: attenuation, shadowing and multipath (Andersen et al. 1995; Hashemi 1993). Attenuation is caused by free space loss, absorption by foliage, indoor partitions and so on. Shadowing (the slow fading) is due to the obstacles between the transmitter and receiver. Multipath fading is caused by propagation on multiple paths and produces short term variation in received power due to the di erent phases on the di erent paths. In order to compute signal attenuation as radios move, it is of interest to study the correlation of the received signal power as the receiver moves from its original position. Two kinds of correlations can be observed in received power: (1) large scale correlation in shadowing, and; (2) small scale correlation in multipath fading. For large scale transmitter-receiver separations, shadowing conditions persist over the range of the physical obstacles which cause the shadowing (Gudmudson 1991). Within the small scale range, say, one half of wavelength to two wavelengths, spatial and temporal correlations between path variables have been observed (Rappaport et al. 1991; Hashemi 1993).. 2.2 The Impact of Channel Characteristics on 3.1 Shadowing Networking Algorithms. Current network algorithm platforms generally assume free space attenuation, which only depends on transmitterreceiver distance. However, the mobility of users through the network system and, consequently, the time-varying multipath channel and interference, could cause the failure of the links and could severely decrease the capacity of the network with respect to the static model. For example, the impact of channel fading on the wireless system can be readily observed through its e ect on link quality (stability), connectivity change and routing table maintenance. The e ect of fading and mobility on network protocols will be accounted. Shadowing, the slow fading, has been generally characterized in the literature by a log-normal distribution, with a standard deviation that depends on the roughness. For example, for indoor environments, a standard deviation of 5 dB is observed (Rappaport et al. 1991). A general path loss model at a speci

(33) c Transmitter-Receiver separation can be described as PL(d) = PL(d0 ) + 10 n log10 ( dd ) + X (1) 0. where PL(d) is the mean path loss in DB at a T-R separation d meters, PL(d0 ) is the free space path loss at a reference dis-.

(34) A SPREAD SPECTRUM DS/CDMA CHANNEL FADING SIMULATION ENVIRONMENT FOR MOBILE RADIO NETWORK SYSTEMS Jack Tzu-Chieh Tsai, Eric Hsiao-Kuang Wu and Mario Gerla Computer Science Department University of California, Los Angeles Los Angeles, CA 90024. Abstract A channel fading simulation environment for wireless networks is presented. The simulation utilizes and extends the SIRCIM statistical impulse response model. For the evaluation of mobile wireless networks, it is important to use a realistic channel model which accounts for the channel correlations between node movements in addition to channel attenuation statistics when the node is static. In this paper, we describe a technique to generate correlated impulse response between di erent positions, and introduce a simulation tool which implements this technique. Numerical examples illustrate the e ectiveness of the simulator and show the impact of accurate radio channel modeling on the performance of mobile radio networks.. 1 Introduction Channel modeling simulation tools that enable researchers and designers to accurately predict the performance of wireless systems become increasingly important as personal communications and wireless data services evolve. A basic understanding of the channel is important not only for designing modulation and coding schemes for robust communication over such channel, but also for investigating the channel fading impact on existing networking algorithms, such as routing and power adjustment which critically depend on channel attenuation. At present, most network protocol simulations and even power control algorithms are using the free space (distance) channel propagation model which is basically only function of transmitter-receiver distance. Typically, for the indoor environment, the channel characteristics are much too complex to be modeled by simple distance functions. Yet, a realistic channel model is essential for network protocol evaluation, especially in the presence of mobility. Therefore, a more realistic channel fading model which accounts for channel quality variations with movement is needed for network protocol simulation. To date, several publications have reported measurements and modeling techniques for indoor radio communications in This work was supported by the U.S. Department of Justice/Federal Bureau of Investigation, ARPA/CSTO under Contract J-FBI-93-112 Computer Aided Design of High Performance Network Wireless Networked Systems. partitioned oce buildings. Among them, a statistical impulse response program, SIRCIM has been derived from empirical measurements (Rappaport and Seidel 1990). The reasons to use statistical channel models instead of incorporating specular re ectors by ray tracing and di raction theory are: (a) the number of potential re ectors is too large to incorporate into a purely deterministic propagation model; and (b) the locations of potential scatterers vary considerably as transmitters and receivers are moved within a building, and from building to building. SIRCIM provides the following models: the distribution of the number of multipath components in a particular multipath delay pro

(35) le; the probability of receiving each multipath component at a particular excess delay; the distributions of the amplitudes, phases, and time delays of multipath components received within a local area. Correlation between multipath component amplitudes for a given excess delay and small receiver separations, as well as at small time delay differences for the same receiver location, are also modeled from the empirical data in (Rappaport et al. 1991). SIRCIM provides channel impulse response at the signal level which is suitable only for radio designs. For network performance evaluation purposes, we are more interested in received power at the packet level. To this end, assuming that the channel is a Direct Sequence (DS) Spread Spectrum channel, we derive the mean signal power by performing convolution of the spread spectrum random chip sequence with the impulse response of the simulated channel. Furthermore, in a mobile radio environment, we must account for the continuity of received power as such power changes because of a change in position. Clearly, random resampling at each position would not provide this continuity. Thus, the received power estimates obtained from the statistical models for adjacent locations must be somehow correlated. To this end, the correlation between received power at di erent locations must be computed. SIRCIM provides only small-scale spatial correlations (in the order of 1 meter). We need to account also for large-scale correlation (up to 10 meters), in addition to small-scale correlation, in order to improve computation eciency and avoid frequent resampling discontinuities. In this paper, we develop a channel simulation tool which is based on SIRCIM models, and which generates correlated channel impulse response between old and new positions after.

(36)