Appropriate Occasion To Trigger TDM Solution
4. DRX parameter:
4.4 SNR and SINR estimations
4.4.4 Numerical examples
Parameter value
Simulation time 140ms
Total system bandwidth 20MHz(100RBs) Block size for time period 2.5ms(5RBs)
Coherence time 3.8ms
Block size for frequency bandwidth 180kHz(1RBs)
Coherence bandwidth 500kHz
LTE center frequency 2.39GHz
UE velocity 30km/hr
Table 4.2: Simulation assumption.
From the simulation result of Fig. 4.21, the channel gain is varying with different symbol time on the same subcarrier. As shown in Fig. 4.22, we assume that the channel gain is varying with different symbols and subcarriers. We notice that the variation of LS+linear interpolation will be serious, but that model based channel estimation will be smooth and much close to the real channel.
Fig. 4.23 shows SNR vs BER curves for different channel estimation methods. The curve from model based channel estimation is always more close to curve of perfect channel than LS estimation method. The advantage of model based method is obvious when user operates at high SNR. The channel estimation error will dominate the bit error rather than noise.
59
Figure 4.21: Channel gain for different symbols.
The model based channel estimation after data decision (Data Aided(DA)-ModelBased method) will be the most accurate method to estimate SNR which is investigated in Fig.
4.24. Only using pilot to estimate SNR (Only Pilot (OP)-ModelBased) will induce over-estimated SNR. Although the pilot position will provide better channel estimation result, noise variance calculation will be underestimated because the number of samples is not enough. The LS channel estimation (DA-LS) will be underestimate seriously when SNR is high. When SNR is high, and the noise variance is low, the channel estimation error will dominate the accuracy of SNR estimation. In that way, it will overestimate the noise variance and make the SNR estimation result lower than the real SNR. When interference becomes serious, and SNR is low, the estimation of SNR will be close to 0dB, but real SNR may be smaller than 0 dB. The accuracy of SNR estimation will be limited by noise because the channel estimation error will contribute to both the signal power and noise power measurement.
Figure 4.22: Channel estimation over frequency and time domain.
Figure 4.23: SNR vs BER for different channel estimation method at UE velocity of 30km/hr.
61
Figure 4.24: Compare different SNR estimation method with real SNR at UE velocity of 30km/hr.
Figure 4.25: Compare NMSE for different SNR estimation method at UE velocity of 30km/hr.
Figure 4.26: Compare different SNR estimation method with real SNR at UE velocity of 80km/hr.
Fig. 4.25 depicts the normalized mean square error performance for different SNR estimation methods when UE velocity = 30km/hr. The MSE for DA-LS and OP-ModelBased both will decrease when SNR is smaller than 5dB, and increase when SNR is larger than 5dB. When SNR is high then the noise variance will be low, the channel estimation error will dominate the accuracy of SNR estimation. Because channel esti-mation in Model based method is more accurate than in LS method, the NMSE will be much less in Model based method. The OP-ModelBased method always get higher SNR result, so the NMSE will be larger than other two methods.
Fig. 4.26 indicates that the model based channel estimator is unreliable because the channel varies fast, and two order regression model is not good enough to chase the channel variation when the block size is 2.5ms in time and 180kHz in frequency. The dopper shift will be Ds = fcc·v = 2.39GHz·80km/hr·10/36
3×108m/s = 177Hz. The coherence time will be 4×D1
s = 1.412ms which is smaller than the block size in time. Hence, the SNR estimation will be inaccurate specially on high SNR because of channel estimation error.
63
Figure 4.27: Compare NMSE for different SNR estimation method at UE velocity of 80km/hr.
The mean square error performance for different SNR estimation methods when UE velocity = 80 km/hr is plotted in Fig. 4.27. Most of the property is approach to Fig. 4.25, but the NMSE increase fast than UE at slow speed because two order regression model is not good enough to chase the channel variation. The OP-ModelBased method will be more precise than other two methods when SNR is high because channel estimation is accurate at pilot. And average SNR at pilot will close to real SNR in enough block size at UE high speed because the small number of time slot will obtain enough statistic property.
From Fig. 4.28, we find that the UE velocity is too fast for the model based channel estimator to have decent performance. We can adjust the block size in time which is 2ms (4RBs) to be smaller. The SNR estimation accuracy will be better than DA-LS.
Fig. 4.29 shows that as the block size in time is small, DA-ModelBased method will be able to track the channel variation, the NMSE will improve. But the NMSE of OP-ModelBased method will become larger than in Fig. 4.27 because the number of time slot is not enough to obtain real SNR statistic property.
Figure 4.28: Compare different SNR estimation method with real SNR at UE velocity of 80km/hr and block size of time is 4(RBs).
Figure 4.29: Compare NMSE for different SNR estimation method at UE velocity of 80km/hr and block size of time is 4(RBs).
65
The above simulation results indicate that the accuracy of SNR estimation for the regression model based approach of [3] is usually better than conventional least square (LS) based channel estimator.