INTERFERENCE REJECTION PERFORMANCE AS A MEANS OF FREQUENCY OPTIMISATION IN A MIXED CELLULAR/MANET NETWORK

Transcription

1 ITERFERECE REJECTIO PERFORMACE A A MEA OF FREQUECY OPTIMIATIO I A MIXED CELLULAR/MAET ETORK Kayonne ebley Faculty Avisor: Dr. Richar Dean Department of Electrical an Computer Engineering Morgan tate University ABTRACT Research at Morgan tate University shows a means of enabling both a mobile a-hoc network (MAET) an a cellular network to operate simultaneously in the same spectrum. This enhance frequency efficiency woul facilitate the creation of a hybri or Mixe Cellular/MAET network (MCM) in which each of the MCM sub-networks woul have access to the entire allotte spectrum. Interference rejection an excision have been ientifie as a means of istinguishing between an isolating the two ifferent kins of signals. This paper shows the promising performance of such techniques within the MCM environment as a part of the integrate etwork Enhance Telemetry (iet) project. KEYORD cellular, spectrum efficiency, iet, MAET, MCM, excision ITRODUCTIO Research is currently being carrie out at Morgan tate University in combining cellular an mobile a-hoc network (MAET) technologies to create a Mixe Cellular MAET etwork (MCM) as part of the integrate etwork Enhance Telemetry (iet) project. The MCM woul provie bi-irectional communication between test articles (TAs) an groun stations (Gs) in test environments. The MCM woul be comprise of a single cell-site borere by one or more MAETs, with the G being irectly connecte to the cellular base station. All TAs that fall within the cellular networks range woul therefore be able to use the cell channels to communicate irectly with the G. The MAET(s) woul be comprise of TAs beyon the range of the cellular network. Data being sent to an from such a TA woul be route through neighboring MAET noes to a gateway noe which woul then sen the ata over the cellular network to the G []. The aim of this paper is to evelop a technique that woul allow the cellular an MAET components of the MCM to operate within the same frequency range. This woul allow iet to use their entire banwith for cellular communications, while

2 simultaneously using their entire banwith for MAET communications as well. pectral efficiency woul therefore be increase with ata rate, capacity an other banwith relate parameters of the network components being maximize. The challenges face involve being able to separate the MAET signals from the cellular signals as neee, as shown in Figure, as both types of networks woul be communicating over the same range of frequencies. This ability woul be especially important for TAs at or near an overlap between the cellular an MAET networks. This paper looks at the performance of perfect filtering an the infinite clipping metho of interference rejection for signal separation, assuming one network uses narrowban TDMA signals,, an the other wieban D signals,. + + RECEIVER Infinite Clipping Perfect Filtering ^ ^ Figure : Role of Gateway oe BACKGROUD Correlation, cross-correlation, autocorrelation an their respective measures were the primary concepts use in analyzing the performance of the interference rejection techniques. For eterministic signals such as the TDMA signals consiere in this thesis autocorrelation is efine as: R x x( t ) x( t ) t, for - < < () where x(t) is a real value energy signal. For ranom signals (an other ranom variables) such as the pseuo-ranomly sprea wieban signal, the expecte value operator is use, such that: R X ( t, t2) X ( t ) X ( t 2) (2) where the ranom variables X(t ) an X(t 2 ) are obtaine by taking time samples at t an t 2 from the ranom signal X(t). The cross-correlation function of two ranom signals x(t) an y(t) is efine in [2] as being: x y p ( xy t, t2 X ( t ) Y( t 2) t t xt, yt ) t () 2 2 The more alike two signals are, is the more correlate they will appear to each other. The crosscorrelation coefficient then can be use for our purposes as a measure of how much two signals interfere with each other, or how ifficult they are to separate. Recall that the primary goal of this work is to enable a TA at the gateway to be able to istinguish between signals that it receives from the MAET an signals it receives from the cellular network. Recall also that both these sets of signals will be operating within the same spectrum of frequency an so will interfere with each other. In orer for the TA to be able to separate out the two, it will have to implement an interference rejection or an excision technique epening on the multiple access scheme that the esire signal is using. If the signal of choice is narrowban an the interfering signal sprea spectrum, then a filter can be use to isolate an 2

3 preserve the narrowban signal while removing the interfering sprea spectrum signal. If the signal of choice is sprea spectrum an the interfering signal is narrowban then an interference rejection technique such as infinite clipping can be use to remove the interfering narrowban signal. Infinite clipping is a special case of excision which, while looking nothing like a filter, actually functions very much like one. It oes this by normalizing the magnitue of the mixe signals to some constant value such as unity or infinity, while preserving the iniviual phase of each signal present. It oes that by multiplying the magnitue of each of the signals by the reciprocal of its magnitue. The magnitues of the interfering signal an the sprea signal are equal, hence e-spreaing the D signal at the receiver results in negligible narrowban interference as Figure 2: Infinite Clipping seen in Figure 2. ith an infinite clipping system, near optimal results are obtaine from quite straightforwar computation. APPROACH It was ecie to use infinite clipping to remove narrowban interferers (BI) when the wieban signal was the esire signal (hereafter referre to as wieban excision), with filtering being use to remove wieban interferers (BI) when the narrowban signal was the esire signal (hereafter referre to as narrowban excision). All performance evaluations both experimental an theoretical - were base on simulations that were carrie out using MATLAB 6.0. All excision operations were one in the frequency omain. This was implemente by carrying out a Fast Fourier Transform (FFT) operation on each of the signals/systems analyze. In both the narrowban excision an wieban excision scenarios, it was assume that interfering signals from the other network were the primary sources of interference with all other interference an noise contributions being negligible. Therefore, the signal to interference ratio (IR) is equal to the signal to noise ratio (R). Performance was measure by consiering the value of the signal to noise ratio (R) at the receiver after separation of the receive signals. ieban excision performance evaluation was carrie out by first creating a system compose of a wieban signal an a varying number of uniform with narrowban interferers. Recall that within the confines of the MCM scenario, the BI are actually non-spurious information bearing signals from the TDMA or FH network. The uniform with assumption is therefore a reasonable one. For the theoretical case, a signal/interferer system in which both the signal an the interferer(s) have been clippe to the same power spectral ensity was assume. The processing gain of the wieban signal as compare to the narrowban signal was taken to be the ratio of the wieban signals banwith to the narrowban signals banwith []

4 B PG (4) B The value of the processing gain was therefore also equal to the maximum number of narrowban signals,, that coul be supporte by the frequency range over which the wieban signal was sprea. ote that since the wieban signal was assume to be sprea over the entire spectrum allotte the MCM network, the processing gain in this case was also equal to the maximum number of narrowban users that coul be supporte by the MCM. If the wieban signal was to be supporting ifferent narrowban signals each over a ifferent frequency range with each range being a subset of the overall wieban signal range, then that range coul also be seen to be ivie into ifferent frequency slots each of BI banwith. If is the maximum number of BI possible, k is the actual number of BI present, an PD is the power spectral ensity of a signal, then in general the R for some value of k can be approximate by ( k) B PD ( k) PD R (5) ( k) B PD ( k) PD where ( k)b gives the banwith of the wieban signal not experiencing interference, an kb gives the entire interference banwith. It was pessimistically assume that wherever interference occurre it completely overwhelme the wieban signal. ince PD = PD in an infinitely clippe system, the R of the wieban signal after infinite clipping is given by: k R k (6) This formula was use to calculate the R for increasing values of k from k = 0 until some maximum k was achieve (i.e. the spectrum was completely fille with interferers). The plot of the relationship between R an k is shown in Figure 4. In the experimental case the wieban signal was simulate by creating a 2 length P sequence. The sprea signal was then assume to have unergone binary phase shift keying (BPK) moulation at the baseban. The Fourier transform of the sprea impulse was taken thus representing the signal in the frequency omain. arrowban signals of uniform with but ranom magnitues were also create, with the number of BI being increase for each successive iteration until the case in which there were k interferers was achieve. For each value of k, the wieban signal an the BI(s) were both clippe to unity PD. This was one by simply iviing each sample of both the BI signal(s) an the wieban signal by the absolute value of itself. This resulte in each signals magnitue being normalize, with the phase remaining unchange. The output of a matche filter can be represente by the correlation of the receive signal with the original signal sent, while the autocorrelation of a real value energy signal with an un-shifte version of itself (T=0) gives the value of the average power of the signal. The matche filter receiver output was therefore represente by the correlation of the receive signal with the original wieban signal. The signal power was calculate by taking the ratio of the zero th value of the autocorrelation of the original wieban signal with the segment of the receive wieban signal that ha not been corrupte by interference (the zero th value of the receivers signal output). The noise power at the receiver was calculate by taking the zero th value of the autocorrelation of the original wieban signal with the BI (the zero th value of the receivers noise output). The R was taken to be the ratio of the former over the latter. The relationship between k an the R receive after infinite clipping was plotte (see Figure 4). 5 4

5 Performance evaluation of narrowban excision was carrie out similarly with simulations being one to obtain both theoretical an experimental results. In this case it was assume that there was only one significant wieban D interfering signal. Experimental R values were calculate by first creating a wieban signal an a narrowban signal. The signal power at the receiver was taken to be the zero th value of the autocorrelation of the narrowban signal with itself, while the noise power was taken to be the zero th value of the correlation of the narrowban signal with the wieban signal. The R was taken as the ratio of the former over the latter. The next step taken was to analyze gateway excision performance as a function of istance. Recall that in orer for a TA to operate as a gateway, it has to be receiving at least a minimum R from each of the two types of network that make up the MCM. Recall that signal power ecreases as the receiver istance from the transmitter (T-R separation) increases. The relationship between receive signal power an T-R separation can be etermine using a path loss moel. In this work, the log istance path loss moel [4] was assume, where n n 0 Pr Pa t PK t (7) where is the T-R separation istance, P t is the transmitte power, P r is the receive power, n is the path loss exponent, a is a constant that has to o with the signal wavelength an antenna performance, an o is the close-in reference istance [4]. For our case o is also consiere to be a constant an so can be groupe with a to form a new constant K. Given the two signals P R P T K n an P R P T K n, representing the narrowban an wieban signals respectively, the following assumptions were mae: - K = K (this assumes that the performance of the respective transmitting antenna from the MAET sub-network is essentially equivalent to that of the transmitting antenna from the cellular sub-network. It also assumes an equivalent energy per bit.) - Path loss exponent: n = (The path loss exponent, n, is equal to 2 in the free space environment an is 4 in the groun base/urban environment. All the noes in the MCM setting being analyze are aeronautical in nature, thus while the MCM environment is not as forgiving as the free space environment it is not as harsh as the urban environment. Assuming a path loss exponent of n = therefore represents a compromise between the two. ) In the wieban excision case, the relationship between receive R an istance can be etermine as follows. Given the signal/interferer system shown in Figure a, note that (a) (b) (c) Figure : (a) ignal/interferer ystem (b) Infinitely Clippe ystem (c) Perfectly Filtere ystem 5

6 PD B K an T 4 PD B K. Assuming that the signal/interferer system is T 4 infinitely clippe, then PD = PD. e assume the presence of k of a possible bans of interferers. ince K = K = K, R is given by: KPD T B k (8) B k k PG B k k KPD T B where PD B T gives the power of the interfering signal after infinite clipping. The relationship between R an istance of a noe from transmitters in both networks ( an ) was plotte. This was one for the values of k that woul result in first % then 66% then 00% of the frequency range occupie by the wieban signal (i.e. the entire spectrum allotte the MCM) being also occupie by BI. Experimental performance values were calculate by using the average after 000 iterations of the experimental wieban excision R values for the %, 66%, an 00% interference occupancy scenarios. For each case, the respective average R value calculate for that percentage of interference was use to replace k in k the above equation. The relationship, therefore, change to become: PG R k (9) AVG This formula was use to plot values of (/) for each of the three values of k mentione above. In the narrowban excision case, the relationship between receive R an istance can be etermine as follows. Given the original signal/interferer system use in the wieban excision scenario, we see that for perfect filtering R is given by: K. PDT B (0) K. PDT B It is reasonable to assume that the original narrowban transmit power is equal to the original wieban transmit power, so PD T B = PD T B. Equation can therefore be simplifie to: K. PD K. PD T T B B PG The relationship between R an istance of a noe from transmitters in both networks (expresse as a ratio of the two istances an ) was plotte. In orer to obtain experimental results, the average R value foun after 000 iterations was use to replace PG in the final form of Equation 2 above. The experimental relationship was therefore given by () 6

7 R AVG A plot of this relationship was create. (2) The final segment of work was to etermine the region of gateway operability. Recall that the minimum require R can be calculate. Accoringly, if a TA is simultaneously receiving a post interference rejection R an R that are both greater than the minimum acceptable R then that TA will be esignate a gateway. If only the R or the R is higher than the minimum acceptable R, then that TA will be esignate to belong to whichever network has the above R min value. If a TA is receiving R an R such that both R's are lower than the minimum acceptable value, then that TA will be sai to be out of range of the MCM. REULT AD AALYI e will first look at wieban excision performance as a function of the number of BI. Figure 4: R of ieban ignals versus Percentage Interferers Figure 4 shows the relationship between the R an the number of BI for both the theoretical an experimental case for the scenario in which infinite clipping was use to remove unwante narrowban signals from a wieban signal. 00% interference inicates that every signal sample is being interfere with. ote that contrary to the norm, the experimental results are better than preicte by the theoretical results. This is because in making the theoretical calculations, the pessimistic assumption was mae that, in essence, the wieban signal an the BI were completely correlate wherever the interference occurre. The more correlate two signals are the harer it is to separate them. Therefore, when complete correlation was assume, the contribution mae to the R by the interference increase causing the overall value of the R to ecrease. Recall that because it has been sprea by a P sequence the wieban signal was, in effect, pseuo-ranomly istribute hence the likelihoo that it coul be completely correlate to the narrowban signal woul be negligible. In the experimental case, therefore, when the actual values for the correlation between the signal an the noise were use to calculate the R the contribution mae by interference was less than preicte an so the R values were higher than preicte. Calculations showe that a minimum R of 0.5 B woul require to support a maximum probability of bit error (P e ) on

8 ote that infinite clipping performance eclines as the number of BI increases. However, the infinite clipping performance is sufficient to prouce a R over 0.5 B when there is up to about an 84% interferer cell loaing in the theoretical case, an 00% interferer cell loaing in experimental case. Finally, recall that this moel assumes an interference limite system (i.e. R = IR, where interference is solely as a result of signals coming from the competing MCM sub-network) where all other noise contributions are consiere to be negligible. Consiering internal noise as a limitation woul result in a reuction of the very high Rs shown here. ignificant reuctions are not, however, expecte. e will next look at wieban an narrowban excision performance as a function of istance. The theoretical relationships foun between the R an the ratio of the istance that a TA is between a transmitter in the wieban network an a transmitter in a narrowban network for both the theoretical an experimental cases are shown in Figure 5 below. Figure 5: Relationship between Excision Performance an T-R eparation Distance - (a) arrowban Excision (b) ieban Excision ote that the experimental results are once again more optimistic than the theoretical results. ote that as the ratio of / increases the receive R after wieban excision also increases in both the theoretical an experimental cases. This makes sense, as the closer a TA is to a wieban transmitter is the greater the wieban signal power that the TA woul receive an the smaller the receive BI signal power (T-R separation istance is inversely proportional to signal power at the receiver). ince in wieban excision the wieban signal is the esire signal, that increase in wieban signal power translates into an increase in receive R. The opposite is true for the receive R. e see that as / increases the R ecreases. This too is logical because the narrowban signal is the esire signal. Hence moving closer to the wieban transmitter an further away from the narrowban transmitter results in an increase in noise power as the signal power ecreases. This translates into a lower R at the receiver. Recall that the gateway requires a minimum R of about 0.5 B from each MCM component in orer to maintain an acceptable probability of bit error. In the narrowban excision scenario we see that R performance meets or excees that minimum for values of / greater than or equal to about -4.8 B in the experimental case an about -2.2 B in the 8

9 theoretical case. In the wieban excision scenario the relationship between R performance an istance epens on the percentage of the spectrum that is aversely affecte by BI. hen the spectrum is about % fille, minimum R performance is met or exceee at ratio values of about B an 4. B for the experimental an theoretical cases respectively. hen the spectrum is about 66% fille, minimum R performance is met or exceee at ratio values of about 9. B an 2.8 B for the experimental an theoretical cases respectively. hen the spectrum is about 00% fille, minimum R performance is met or exceee at ratio values of about 6.7B an 2. B for the experimental an theoretical cases respectively. The final segment of the work seeks to efine the straight-line range over which a TA can travel an remain capable of being a gateway noe. The experimental values calculate for the gateway region are more realistic than the theoretical values calculate. In the experimental case a R of 0.5 B or greater from both networks can expect to be supporte within a range of - 4.8B / 6.7B when up to about 00% of the wieban signal is affecte by the narrowban signals, -4.8B / 9.B for up to about 66% narrowban signal interference, an -4.8B / B for up to about % narrowban signal interference. ( w / ) min = -4.8B ( w / ) max = 6.7B Cellular A Hoc ( w / ) min = -4.8B ( w / ) max = 9.B ( w / ) max = B ( w / ) min = -4.8B Figure 6: Range Capable of upporting Gateway Operation (Experimental) A gateway woul therefore be able to operate within these ranges, as epicte in Figure 6. The results suggest that there is a wie region over which a TA woul be able to serve as a gateway (by satisfying the 0.5 B minimum R requirement). ote also that this woul be possible even in conitions in which the cellular component of the MCM is operating at, or near, full occupancy. COCLUIO The following conclusions were mae from the results foun: - Given a particular moulation scheme an minimum acceptable P e, it is possible to calculate the minimum R that a receive signal nees to have. hen BPK is use an a P e of at least 0 6 is require, the minimum acceptable R at the receiver is approximately 0.5 B - It is possible through infinite clipping an filtering techniques to have the MAET an cellular components of the network operate on the same frequency range with satisfactory R. 9

10 These calculations were mae assuming a single wieban signal. BI can be separate from wieban interferers up to about an 84% loaing of the narrowban component of the MCM in the theoretical case, an over 00% in the experimental case. - In the propose MCM it is possible to etermine which network a TA shoul be in by monitoring the R it receives from both components of the MCM an comparing it to the minimum R which in this case is equal to 0.5 B - Gateways can expect to have a reasonable range (in terms of istance) over which to operate, from a minimum of a ratio of istances of -4.8B up to B (given by / ). REFERECE [] K. ebley, A Look at Cellular Packet Data Performance for Application in iet, in Proceeings of the International Telemetering Conference, Las Vegas, V, October 2005 [2] J. G. Proakis, Digital Communications, 4 th Eition, Mcgraw Hill, Y, 200 []. tallings, ireless Communications an etworks, 2 n Eition, Pearson Prentice Hall, Pearson Eucation Inc., Upper ale River, J 07458, 2005 [4] T.. Rappaport, ireless Communications: Principles an Practice, 2 n Eition, Pearson Eucation 0