Characterising mobile WiMAX performance

This article describes how a channel emulator can be used to characterise the performance of a MIMO receiver. The testing was done in stages of increasing complexity, namely, testing under AWGN conditions, MIMO testing with known static channels and, finally, testing with channels chosen to represent ‘real world’ behaviour.

The article seeks to demonstrate how testing at each of these stages can help to give engineers confidence in their design as well as potentially expose issues which may be difficult to isolate with the more complex ‘real world’ testing.

The IEEE 802.16 working group is focused on developing standards for broadband wireless access. As part of this work it has written a physical layer specification for mobile devices that uses orthogonal frequency division multiple access (OFDMA) in combination with techniques such as adaptive modulation and coding (AMC) and multiple input, multiple output (MIMO).

The overall performance gains for a network that uses techniques such as AMC and MIMO are strongly influenced by the receiver characteristics of client devices within that network.

For example, a good receiver will allow the network to more readily use complex modulation schemes and, hence, achieve greater throughputs.

Figures 1 and 2 show a schematic representation and a picture of the equipment used for the testing.

Figure 1: Equipment set-up for measuring receiver performance.

Figure 2: Measurement set-up used for receiver testing.

The E6651A is a fully functional base station emulator designed for testing mobile WiMAX devices. It supports a number of different use models, including RF testing, functional emulation of protocol features, end-to-end application testing and protocol conformance testing.

In this case the equipment was used to perform a downlink ‘ping test’, with a variety of different modulation and coding formats both with and without MIMO.

The test set has two RF ports, that can be configured to provide transmitter outputs for MIMO testing. Each downlink signal was fed into a signal analyser (MXA) which downconverted the RF signal to a differential baseband digital signal which served as an input to the PXB.

The N5106A PXB MIMO receiver tester is a baseband channel emulator that allows the user to emulate a variety of single channel and MIMO fading conditions. In addition to channel characteristics, such as Doppler spread and delay spread, the PXB can emulate antenna characteristics, for example antenna spacing, polarisation, antenna lobe patterns and angular spread.

The PXB can be used in a baseband configuration where it is connected to a signal source (MXG) and used to test receiver performance for a variety of different wireless formats.

In this test set-up the digital outputs from the PXB were up-converted to RF using a signal source and then connected with the device under test. To support a fully functional radio, the cabling set-up uses isolators to provide an unfaded uplink signal to the RF1 port of the E6651A.

Figure 3 shows sensitivity measurements for the device under test (DUT) under thermal noise conditions. The figure is also annotated to show the test limits as defined by the WiMAX Forum in its Radio Conformance Test (RCT) specification.

Figure 3: Packet error rate under WGN conditions.

These results demonstrate that the receiver is meeting the required sensitivity limits under AWGN conditions. The WiMAX Forum test limits are set by calculating the theoretical thermal noise power and then assuming degradation due to the noise figure of the receiver and an implementation loss.

Passing this test provides a level of confidence about both the device and the measurement set-up.

MIMO techniques exploit the multipath characteristics of radio channels and allow the link to either benefit from higher orders of diversity or to create separate spatial streams that can support the transmission with increased data throughput.

The use of spatial streams for increasing data throughput is a development which, given the right conditions, provides the possibility of achieving data rates that exceed the Shannon Capacity limit.

Given that MIMO is exploiting the characteristics of a multipath channel, it is not possible to characterise the performance using a simple AWGN channel. However, before testing the performance in channels which are intended to emulate ‘real world’ environments, it is helpful to test the MIMO performance in simple static channels where the expected behaviour is known.

One way of characterising whether or not a ‘channel’ is well suited to providing multiple spatial streams is to calculate the condition number, a measure of how sensitive the eigenvalues of a given matrix are to small perturbations of the values in that matrix.

If the condition number is low then the eigenvalues will not be sensitive to small perturbations (or noise) and the channel will be suitable for supporting two MIMO streams. A high condition number indicates that the channel matrix is very sensitive to noise and the system requires an increased signal to noise ratio to support multiple spatial streams.

Figure 4 shows a schematic representation of a 2x2 MIMO channel. Each of the Tx-Rx pairs has its own channel which, in the general case, can be represented by a complex time varying impulse response.

Figure 4: A 2x2 MIMO channel.

For the simplified static channels, h00, h01, h10 and h11 can each be represented by a constant complex number within a single matrix. The matrix (1,0,0,1) has a condition number of 0 dB and reflects a theoretically perfect MIMO channel with two spatial streams that are orthogonal.

The packet error rate curves in Figure 5 show a shift in the required SNR for a single transmitter, single receiver (SISO) link vs an ideal MIMO link with two transmitters and two receivers (condition number = 0 dB).

Figure 5 also shows curves for packet error rates for a variety of channels with varying levels of condition number.

Figure 5: MIMO performance in static channels with varying condition numbers.

Figure 6 shows an example where the blue line corresponds to the additional CNR required to sustain an EVM of 32%.

Placed on this graph are three data points (shown as a yellow circle, square and triangle) which overlay the measurements presented in this article. This shows a good correlation between the SNR required for a given EVM vs those related to specific bit error rate curves.

Figure 6: Increased SNR requirements vs condition number.

Figure 6 also shows that, for a practical receiver implementation at high condition numbers, MIMO techniques fail to provide a useable benefit as the system suffers from irreducible error rates (see curve for condition number = 25 dB).

Testing with static channels is instructive as it allows comparison between measured results and the theory of MIMO operation. However, to assess the expected performance of a given system, it is necessary to emulate time varying ‘real world’ channels.

In this article, an ITU pedestrian profile B (mobile speed of 3 km/h) was used as one example of a real world channel.

Figure 7 shows the SISO receiver performance for both ½ rate QPSK and 5/6 rate 64 QAM under both AWGN and pedestrian B channel conditions. For the ½ rate QPSK, the sensitivity is degraded by around 8 dB, using the pedestrian B channel at the 1% PER level. For 5/6 rate 64 QAM, the receiver suffers from an irreducible error rate when measured under pedestrian B channel conditions. To verify this result the measurement was repeated on a different DUT and similar results were found.

Figure 7: QPSK vs 64 QAM under noise and pedestrian B channel conditions.

Figure 8 shows the 5/6 rate 64 QAM performance with a variety of SISO/MIMO configurations. The green line for single input, multiple output (SIMO) shows the benefit of Rx diversity with the irreducible PER reduced to only 0.1%.

Figure 8: WiMAX MIMO Receiver performance.

This is expected theoretically if the errors on each of the channels can be considered to be mutually exclusive random variables (ie, 0.01 x 0.01 = 0.0001). Interestingly, if Matrix A MIMO is employed (ie, transmit diversity), the irreducible error rate is 1% and not 0.01% achieved with Rx diversity.

This could be explained by quantisation issues during deep fades as, for Tx diversity, each Rx has to extract the two Tx signals from one sampled waveform, which is not required in simple Rx diversity.

Matrix A MIMO does, however, demonstrate a diversity gain versus the blue SISO curve. Given that WIMAX networks seek to optimise around a 10% PER, the Matrix A performance gain will be valuable to the system.

Finally, Matrix B MIMO demonstrates that, even under these Ped B channel conditions, it is possible to double the data throughput. There is, however, a penalty in terms of the required signal to noise.

This article shows how a base station emulator can be used in conjunction with a channel emulator to characterise the performance of a MIMO capable receiver. Measured results were compared with theory for both AWGN and static MIMO channel conditions.

Finally, this article presents MIMO receiver performance results using the ITU pedestrian B profile. This work shows how the ITU pedestrian B profile does stress the receiver (witnessed by the presence of an irreducible error rate), but that it is still possible to exploit the benefits of both receive diversity and MIMO in this radio environment.