Results and Contributions
We built a detailed indoor propagation model using ray tracing in Matlab. The reflection and transmission attenuation data from [langen94] was incorporated into the model. For any given room configuration, the model provides accurate signal strength information for any pair of transmitter and receiver locations. The model includes the LoS path and the first order reflections from all reflective surfaces. Therefore, we also obtain accurate signal strength data for total interference at all points in the room for any arbitrary set of transmitters. Figure 5(a) shows some of the first order reflections in an empty room where the transmitter is the access point at the center of the ceiling. The propagation model we developed is used for all of our experiments.
In order to maximize data rates in our architecture, we allow multiple regions in the room to be active at the same time. Therefore, nodes will see interference from three sources:
- Interference from the side-lobes of the beam formed by the antenna towards other regions.
- Interference from the first order reflection of the desired signal that arrives later than the symbol time (2.5ns). Figure 5(b) in red shows this inter-symbol interference while the white shows the reflection that arrives within the symbol time and hence contributes to total signal strength.
- Interference from the first order reflection of all the other simultaneous transmissions.
We use three techniques to combat interference:
The distance of the access point to different parts of the room ranges from 3 to 8 meters. Therefore, we use different transmission powers in different regions to reduce interference to other regions. The transmission power (in dB) is computed using the following link budget calculation:
Sensitivity = Noise + Bandwidth + RXnoise + SNRmin
where, RXnoise - noise figure in the receiver circuits and SNRmin - minimum Signal to Noise Ratio requirement
If we use 64QAM (Quadrature Amplitude Modulation) as an example, to achieve a target BER (Bit Error Ratio) of 10^(-6), we require a SNR_min (Signal-to-Noise Ratio) of 23.4dB for a Rx_noise figure of 10dB and standard thermal noise of -174dBm/Hz. From this we can calculate the minimum required transmit power (in dB) for region i as:
Ptx(i) = max Ptx(d,i) = max(Sensitivity - Gtx(i) - Grx(i) - Path Gain(d,i))
where, Ptx - transmit power, Ptx(d,i) - transmit power required for a receiver at distance d from the AP in region i, Gtx - antenna gain at transmitter, Grx - antenna gain at receiver, and Path Gain - attenuation with distance for this frequency
Figure 6 shows an example of transmit power for the case when using 64QAM with BER requirement of 10^(-6). In our simulations, we considered four different modulation schemes listed below. The modulation used for a given user is the one that maximizes data rate within the BER constraint of 10^(-6).
- We use 1/2 or 3/4 convolution codes (constraint length 8 ) to get an additional gain of between 3-7dB.
- Each antenna module has M=20 elements. We use the M-1 nulling idea described previously to reduce the sidelobe interference.
We use the following table for our experiments.
QAM: Quadrature Amplitude Modulation
Figure 7 shows the data rate obtained as a function of the number of users when we use 7 channels (each 640MHz). The plot labeled “static” corresponds to our STDMA (Spatial Time Division Multiple Access) algorithm while the “dynamic” corresponds to the greedy algorithm where no regions are used and we beamform towards individual users. We see a linear scaling for our algorithm while the dynamic algorithm’s performance falls when there are too many users. The reason was previously discussed in the context of Figure 2(a). When there are 10 users in the system, our approach delivers 1 Gbps/user data rate but this falls to 600Mbps/user when there are 50 users. The reason is increased interference between simultaneous transmissions. However, recall that these numbers are based on simulations for a room size of 10m x 10m. Placing 50 users in such a small room is unrealistic and therefore one can argue that our scheme does provide Gbps/user for realistic room usage scenarios. Another way of looking at the result is that we achieve an aggregate data rate of 30Gbps in the room which translates to 300Mbps per meter square.
In order to study energy scaling with data rate, we varied the number of modules from 1 to 21 for the 10mx10m room. Figure 8(a) plots the energy per bit as a function of the number of modules K. When K is small, we get a low energy cost because there is little interference between simultaneously active regions. As the number of modules increases, more regions are simultaneously active causing higher interference. However, when K is less than 13 the gain in throughput offsets the increased energy cost due to interference. When K is greater than 13, the interference is much more significant resulting in high energy per bit. The reason for this is that by forming M-1 nulls, we force the main beam to shift slightly thus lowering gain in the direction of the desired signal, Figure 8(b). Finally, when K is greater than 15, there are more regions towards which we form only 1 null and therefore the behavior illustrated in Figure 8(b) is less pronounced. This results in higher signal strength in the main beam and the corresponding decrease of energy per bit illustrated in Figure 8 (a).
Figure 8 (a)
Figure 8 (b)
Repairing Links using Reflectors
Links are easily broken by user mobility or by obstructions. As discussed previously, we use wall mounted passive reflectors to repair links. Figure 9 shows the throughput for each region when region 1, 4 and 13 are covered by reflected paths. Note that the throughput for the reflected paths is lower because the path is longer than the LoS paths and the signal is attenuated by 3dB when reflected. Region 13 has the lowest throughput since it is in the center of the room and thus suffers the highest interference in addition to lowered signal strength due to the longer path.
Summary of Contributions
We achieve our goal of delivering Gbps/user data rate under a variety of conditions. The specific research innovations include:
- An innovative M-1 nulling technique to reduce the interference from simultaneous transmitters to the level of thermal noise.
- Algorithm to repair broken links using passive reflectors that provide alternate paths between the access point and users.
- A detailed simulator in Matlab using ray tracing and real beamforming models for this frequency. Accurate propagation models where also constructed utilizing actual measurement data.
- Resource allocation algorithms for highly efficient frequency reuse that exploit the actual beam shapes.
- An innovative new metric for characterizing capacity: bps/meter square.