Deliverable 3

7.2Performance Comparisons among the Candidates

7.2.1SM codes

In this section the performance of SM, eSM and PH-eSM are compared by means of simulations. The new channel model developed under the DVB-NGH standardization process is used. The system parameters chosen for the simulations are illustrated in Table and are the same parameters employed in the standardization process of DVB-NGH for the evaluation of the MIMO schemes.
Figure 10 illustrates BER against SNR for the three schemes under evaluation. For this simulation, the channel model detailed in [80] is modified from its original description, removing the frequency shift for the first line of sight (LOS) tap of the cross-polar term. Moreover the simulation includes power imbalance matrices specified in [80], which model asymmetries between the H/V components of the transmit antennas. Results showed in Figure 10 illustrate similar performance of the three simulated schemes. Due to the similar robustness of the schemes under the simulated scenario, a set of corner cases are defined to highlight differences modifying some of the channel parameters like correlation and Rice parameter K.

Table : Simulation parameters Signal Definition FFT size: 4K Number active carriers: 3409 (no pilots used) GI fraction: 1/4 Constellation: 8 bits per carrier (16QAM + 16QAM) LDPC code: 16K, code rates 4/9 and 2/3 Time interleaver (TI) memory size: 525Kcells Type 1 PLP Channel Helsinki 2 portable outdoor Doppler spread: 33.3 Hz A new realization is generated for each TI block A different sequence of random seeds is used for each SNR point. A same sequence of random seeds is used for all MIMO schemes per SNR point Receiver Ideal channel estimation Simulation stop conditions Minimum simulated TI blocks: 1e3 Maximum TI blocks: 1e4 Erroneous TI blocks: 1e2 Erroneous TI block = at least one erroneous bit

Starting from the scenario described for simulations in Figure , power imbalance matrices due to asymmetries are removed and the TI memory is reduced to 100 Kcells to speed up the simulations. Figure and Figure show BER vs. SNR performance results where the correlation of the standard channel model is increased. For code rate 4/9 PH-eSM outperforms eSM and SM by 0.20 dB and 0.35 dB respectively whereas for code rate 2/3 eSM and PH-eSM have similar performance outperforming SM by 0.26 dB approximately. For Figure and Figure the Rice K factor is increased and the performance order is similar as in the previous case. For code rate 4/9 PH-eSM outperforms eSM and SM by the other two schemes by 0.30 dB and 0.36 dB respectively, finally for code rate 2/3 PH-eSM outperforms eSM and SM by 0.10 and 0.17 dB respectively.

Figure : BER vs. SNR curve for standard case with code rate 2/3	Figure : BER vs. SNR curve for high correlation corner case with code rate 4/9
Figure : BER vs. SNR curve for high correlation corner case with code rate 2/3	Figure : BER vs. SNR curve for high K corner case with code rate 4/9
Figure : BER vs. SNR curve for high K corner case with code rate 2/3

7.2.2Rotated constellation-based Spatial Multiplexing scheme (rSM)

The rotated constellation-based space code proposed by Telecom Bretagne has been simulated and compared with its competitors in DVB-NGH, in particular with the eSM scheme. The simulation conditions are list in. Two simulation campaigns have been carried out. For the first campaign, the “Helsinki 2 MIMO outdoor portable model”, defined in [80] was adopted. The following figures provide some examples of performance results, targeting 16-QAM constellations, obtained by Telecom Bretagne during the first simulation campaign.


_{Table : Simulation parameters}

Parameter	Value
Channel Bandwidth	8 MHz
FFT size	4k
Guard interval	1/4
Carrier frequency	600 MHz
Time interleaver depth	100 ms
LDPC length	16k
Code rates	1/5, 1/3, 4/9, 2/3
Channel estimation	ideal
Time interleaver depth	100 ms
LDPC decoding algorithm	product algorithm, 50 layered iterations
Number of simulated samples	Number of simulated samples: 2000 FEC frames in error for BER > 10^-3, 1000 FEC frames in error for BER < 10^-3

<sub>Figure : BER/FER comparison of rSM and eSM schemes. 16-QAM constellation, Doppler frequency fd =194.8 Hz, coding rate R =1/5.



Figure : BER/FER comparison of rSM and eSM schemes. 16-QAM constellation, Doppler frequency fd =194.8 Hz, coding rate R =1/3.



Figure : BER/FER comparison of rSM and eSM schemes. 16-QAM constellation, Doppler frequency fd =194.8 Hz, coding rate R =4/9.



Figure : BER/FER comparison of rSM and eSM schemes. 16-QAM constellation, Doppler frequency fd =194.8 Hz, coding rate R =2/3.</sub>
The main conclusion of the first simulation campaign was that eSM and rSM shows very similar performance for all coding rates. Thus, a second simulation campaign was carried out in order to decide between the different proposed solutions. During the second simulation campaign, corner cases were agreed to supplement the previously defined NGH “Helsinki 2” outdoor channel model: different values of Doppler frequency, power imbalance (PI), Ricean K factor and cross-polar discrimination (XPD) factor were used. A high correlation scenario was also added.
The following figures illustrates, in the case of 16-QAM constellation and coding rate 4/9, the different corner cases that have been simulated during this second simulation campaign in order to get a winning technique.



<sub>Figure : BER/FER comparison of rSM and eSM schemes. 16-QAM constellation, Doppler frequency fd =33.3 Hz, K = 0, XPD = 2, PI = 0dB, normal correlation matrix, coding rate R =4/9.



Figure : BER/FER comparison of rSM and eSM schemes. 16-QAM constellation, Doppler frequency fd =33.3 Hz, K = 0, XPD = 2, PI = 0dB, high correlation matrix, coding rate R =4/9.



Figure : BER/FER comparison of rSM and eSM schemes. 16-QAM constellation, Doppler frequency fd =33.3 Hz, K = 0, XPD = 2, PI = 3dB, high correlation matrix, coding rate R =4/9.



Figure : BER/FER comparison of rSM and eSM schemes. 16-QAM constellation, Doppler frequency fd =33.3 Hz, K = 1, XPD = 2, PI = 0dB, normal correlation matrix, coding rate R =4/9.



Figure : BER/FER comparison of rSM and eSM schemes. 16-QAM constellation, Doppler frequency fd =33.3 Hz, K = 1, XPD = 2, PI = 3dB, normal correlation matrix, coding rate R =4/9.



Figure : BER/FER comparison of rSM and eSM schemes. 16-QAM constellation, Doppler frequency fd =33.3 Hz, K = 1, XPD = 8, PI = 0dB, normal correlation matrix, coding rate R =4/9.



Figure : BER/FER comparison of rSM and eSM schemes. 16-QAM constellation, Doppler frequency fd =33.3 Hz, K = 5, XPD = 2, PI = 0dB, normal correlation matrix, coding rate R =4/9.</sub>

Although the corner cases were intended to decide between the competing techniques, rSM and eSM still showed similar performance, with very small gaps between the curves. The eSM technique was finally adopted in DVB-NGH as 2x2 code.

7.2.34x2 codes

The performance comparison of the proposed 4x2 schemes to selected 2x2 schemes and SISO and SIMO are shown here. The simulated channel is NGH outdoor channel with Doppler frequency of 33.3 Hz and delay of 0.05 times the guard interval length for the second transmitter pair with respect to the first one. The other system parameters used in the simulations are shown in Table .
_{Table : System parameters used in the simulations}

Parameter	Value
Bandwidth	8 MHz
FFT size	4k
Guard interval	1/4
Modulation	QPSK
LDPC length	16k
Code rates	1/4, 1/3, 2/5, 1/2, 3/5, 2/3
Channel estimation	ideal
Time interleaver	2^18 cells

The SNR vs. bpcu curves are presented in Figure . To obtain the spectral efficiency (bpcu) pilot density of 1/24 for SISO, 1/12 for 2x2 and 1/6 for 4x2 is assumed. All codes are decoded using soft output sphere decoder with max-log approximation. Ideal channel estimation is applied.

_{Figure : Performance comparison of 4x2 MIMO schemes.}

To obtain the capacity curves, LDPC code rates 1/4, 1/3, 2/5, 1/2, 3/5 and 2/3 for all studied SFBCs are simulated and the SNR for reaching BER is searched. Thus, each point in the curves corresponds to SNR required to reach the BER criterion. This BER criterion was selected as a compromise between the simulation time and accuracy. First of all, it is visible that multi antenna techniques provide gain over SISO and SIMO due to increased diversity. Also, rate 1 4x2 codes show good performance at low signal-to-noise levels. Further, rate 2 4x2 codes outperform 2x2 SM (Spatial Multiplexing). It is important to note that altering the pilot carrier density does not affect the order of 4x2 curves, but changes the bpcu point where the capacity curves cross. Also, if 4x2 schemes are compared to 2x2 schemes, having denser pilot pattern, for example 1/12 for SISO, 1/6 for 2x2 and 1/3 for 4x2, begins to eat out the performance gain of 4x2 schemes.