|
7.4. Interference Analysis: NB on UWBIn the frequency range above 3.1 GHz, the FCC mask restricts the radiated power to 41.25 dBm/MHz, with a 10 dB bandwidth of 7.5 GHz. In the 2.4 GHz ISM and 5 GHz NII bands, allowed emissions are 40+ dB higher per MHz. With a minimum required bandwidth of 500 MHz, the UWB receiver will see multiple NB waveforms. The range below 1 GHz, of course, is densely occupied by licensed NB systems (AM, FM radios, TV stations, and so on). The wide-bandwidth of UWB systems permits large processing gains. However, the individual pulses have very little energy; as such, NBI signals, which may be expected to be 1012 dB above the noise floor, may cause severe degradation in performance. The impact of NBI on single-user systems is studied in [28], [29], [35], [37]. A DSCDMA UWB system is shown to resist NBI from IEEE 802.11a OFDM signals [36], where the NBI was modeled as Gaussian noise, as was the MAI. Use of a maximal ratio combiner (MRC) can suppress MAI but not NBI [34]. BER expressions were derived in [38] for a multi-user TH system operating over a pure AWGN channel (that is, no NBI), which could provide a benchmark to study the performance of NBI suppression schemes. Consider the time-hopped UWB waveform described in (7.10). Assume PAM modulation so that D = 0. Then, the transmitted waveform during the j-th UWB symbol is Equation 7.35 The received signal during the j-th UWB symbol interval is Equation 7.36
where wj(t) denotes AWGN with two-sided psd No/2, and ij(t) is interference. In AWGN, the optimal receiver is a MF, matched to the symbol waveform, Equation 7.37
Equivalently, it does the following: it time gates the received waveform, followed by matched filtering with the pulse shape p(t), and then despreading with the spreading code ; that is, the decision statistic is Equation 7.38
Equation 7.39
where Sj, Wj and Ij denote signal, noise, and interference components. Because the pulse p(t) has unit energy, and {ak} is a ±1 sequence, we readily obtain
The noise term Wj is zero-mean Gaussian with variance NfNo/2. The interference term ij(t) represents the aggregate effect of modulated or un-modulated narrow-band signals. Let f, , and denote the carrier frequency, rate, and power of the -th NBI. The received signal component due to the -th interferer can be written as Equation 7.40
where
Because the UWB signal is time-hopped, there is a random time-offset between successive pulses of the same bit. Hence, the starting phase of the NBI carrier changes randomly from the time associated with the beginning of one pulse to that of the next. Given a pulse duration Tp, and an NB symbol duration T = 1/, the probability of a symbol transition within an UWB pulse is given by T. For a UWB pulse of duration 2 ns and a NB signal with a rate of 50 KHz, this probability is 2 -th NBI to the decision statistic where ,k,l denotes the symbol modulating the th NBI during the j-th UWB bit, with ej,k, denoting the corresponding delay. The phase term fj,k, incorporates both the unknown NBI carrier phase as well as the random time shift between pulses due to time hopping. Given that a constant over the duration of the pulse, but this amplitude varies randomly over {j, k, }. For a rectangular pulse shape, = 1. Recall that [10] and P(f) its FT; hence, we have
Equation 7.42
For a symmetric pulse shape, P(f) = P(f) is real-valued and Equation 7.43
The phase variables, the fs, are random variables and are well modeled as being independent (across j, k, l) and uniform over [0, 2p). If q ~ U[0, 2p], we have for m 0, ejmq} = 0, and Var{ejmq} = 1 and phase f are mutually independent, we have E{Ij,} = 0 where the expectation is taken with respect to the fs and bs. Next, we compute the variance: If Nf is large, we can approximate Ij, as a zero-mean Gaussian r.v. with variance . Because the interferers are modeled as independent, their variances will add. Consequently, we can write the decision statistic as Equation 7.44
where hj is zero-mean Gaussian with variance . The BER is then given by Equation 7.45
where SNRp := Ep/(No/2) is the per-pulse SNR, and SNR := /(We can derive (7.45) in a more heuristic, but insightful, way as follows. Assume that the UWB pulse has a flat FT so that , , and that the PSD of the NBI is flat over , where fu and fc are the center frequencies, and W, B are the bandwidths of the UWB and NBI signals. Assume that the support of the NBI PSD is contained within that of the UWB PSD. Then a simple calculation yields the interference power as . With the interference modeled as Gaussian, this leads to (7.45). As a specific example, consider the Gaussian pulse of Example 7.1, p(t) = exp(t2/2s2u). The FT of this pulse peaks at f = 0; the FCC specifications require that the PSD be down by 34 dB at f = 0.96 GHz. To ensure this, we need su (½)109. The time domain signal and its FT are shown in Figure 7.12 for s = (½)109. The pulse width is about 1012 ns, and the 3 dB and 10 dB bandwidths are 0.28 GHz and 0.52 GHz. Figure 7.12. The Gaussian UWB Pulse, and its Power Spectrum.Figure 7.13 shows signal spectra using the Gaussian pulse and the Raised Cosine Pulse (RCF) for different values of symbol period T and excess bandwidth parameter b. These choices of the RCF meet the FCC specs at 0.96 GHz, while utilizing the spectrum more efficiently (ability to transmit more energy). Note also that the Gaussian pulse extends over 1216 ns, the RCF pulse extends over typically 68 symbol periods, so that the RCF pulses have shorter duration. Finally, if the channel has no delay spread, the RCF pulse is desirable because it satisfies the Nyquist criterion. Figure 7.13. A Comparison of the Power Spectra of Gaussian and RCF Pulses.
|
|