Optimizing WLAN performance with MIMO calls for careful analysis

The next-generation 802.11n standard allows for an increase in the number of MIMO radios and other options. As a result, WLAN product developers are beginning to plan product differentiation around such upcoming options.

Most industry observers assume a 2x2 (two-transmit by two-receive) MIMO architecture will be widely deployed. But it turns out that a 3x3 architecture offers highly cost-effective performance benefits for wireless LAN (WLAN) access points.

When the standard is approved it will include MIMO options such as transmit beam forming, space-time block coding, cyclic delayed diversity, maximal ratio combining, and intelligent antenna selection. Standards development has focused most of the attention on these MIMO techniques, and they are quite useful for improving WLAN throughput, range and robustness.

Increasing the number of MIMO transmitters and receivers, however, holds even greater potential. As a simple rule of thumb, the more radios you use, the better your range and throughput. Up to a point, you can obtain this improvement at low cost in terms of both dollars and power consumption.

MIMO architecture tradeoffs
To show the tradeoffs, this article compares the cost/performance of conventional 2x2 MIMO with 3x3 and 2x3 architectures. The latter is a useful option for client devices, where a full 3x3 architecture may be difficult to implement.

All of the MIMO configurations analyzed for this article utilize two unique data streams, no matter how many transmitters and receivers are used. Also bear in mind that throughput and range are closely related in 802.11 WLANs because throughput decreases as range increases.

For this reason, technologies that improve range also improve throughput for WLAN nodes that are farther from the access point. (Channel capacity also affects throughput, and the 802.11n standard will likely include the option of using 40-MHz channels. Legacy WLAN standards all use 20-MHz bandwidths, so 40-MHz mode will double throughput.)

Optimizing the range/throughput relationship is crucial for most applications—especially WLANs for large homes that deliver broadband Internet and digital media content to multiple family members. A significant range/throughput improvement makes it possible for home users to get the same kind of trouble-free service they expect from wired connections.

Thus, such an improvement is not just incremental; it is the key to achieving success for a broad range of WLAN products in the consumer market, especially as applications and media content demand greater bandwidth throughout the home and SOHO environments.

The unexpected MIMO option
Multiple input/multiple output (MIMO) systems use multiple transmitters and receivers to improve performance. Based on theoretical work that began in 1984, MIMO has become practical today through the use of highly integrated radios-on-a-chip. Integrating more radios improves range and throughput, but you have to consider the tradeoffs of cost and power consumption.

To understand the advantages and tradeoffs, it is important to begin with an understanding of spatial multiplexing. This technique enables a wireless system to increase throughput without using additional spectral bandwidth.

The easiest way to understand spatial multiplexing is to think of a system transmitting N unique data streams using highly directional antennas aimed at N different receive antennas (Figure 1). Each receiver detects a unique data stream that travels via a unique physical path, resulting in an N-fold increase in throughput.

Click here for Figure 1

Sophisticated signal processing techniques eliminate the requirement for directional antennas by providing the ability to detect and differentiate multiple data streams—so long as these streams travel via unique paths. This is a case in which multipath effects are actually useful, because they enable signal-processing systems to sort out the different data streams. Further, the more unique paths an environment provides, the more potential there is to increase throughput.

Highly integrated WLAN chips make it practical to exploit these multiple channels in a cost-effective way. But how many channels are realistic for an application such as a home network? The analysis that follows answers this question based on cost, power and range/throughput benefits.

Improving on conventional 2x2 MIMO
The simplest spatially multiplexed MIMO system contains two transmit chains, two receive chains, and two data streams. When expanding such a 2x2 architecture, several factors need to be considered.

In most applications, for example, it is reasonable for the access point (AP) to transmit at a higher power than the client devices, since the latter are often small, battery-powered products. For similar reasons, the AP generally has greater flexibility for the physical placement of multiple antennas. This analysis thus looks at the downlink (AP to client) separately from the uplink (client to AP) direction.

The analysis begins by comparing performance simulations for 2x2 and 3x3 access points (both using two unique data streams). To make these simulations realistic, they include the operating characteristics given in the ITU1238 model (the International Telecommunication Union’s statistical description of typical home operating conditions): 11g band (2.4 GHz), 20-MHz mode, 50-ns rms delay spread, 3.1dB path loss, 15dBm power per transmit chain, 3dB antenna gain, 5dB system receiver noise, 12dB shadowing margin (from multipath), and 4dB floor loss (signal attenuation through a wooden floor).

Figure 2 shows the results of simulations comparing the 3x3, 2-stream system with a reference 2x2, 2-stream configuration.

Click here for Figure 2

The simulations reveal that the 3x3 configuration outperforms the 2x2 configuration in the downlink direction (access point to clients) by better than 40 percent in the 20 to 100 ft range.

Because 3x3 devices are only now becoming available, chamber tests have not yet been completed. However, initial real-environment tests of a recently announced Atheros 3x3 device shows an average 44 percent performance advantage over the 2x2 configuration at 50 feet, 51 percent at 100 feet, and 62 percent 150 feet.

Now consider the uplink direction (client to access point). As mentioned, clients generally do not have as much flexibility to accommodate multiple antennas and radiate more power, so a full 3x3 configuration may not be appropriate for clients. A 2x3 system (transmitters x receivers) can work for clients, however.

Figure 3 shows the results of simulations comparing such a system with the reference 2x2 configuration (using the same ITU1238 model as the previous simulation). As Figure 3 shows, the 2x3 system improves average uplink performance by about 20 percent compared to the 2x2 system. Even though both systems use two spatially multiplexed streams, the additional downlink transmitter significantly improves performance.

Click here for Figure 3

Figure 4 shows the throughput results obtained in chamber testing of the 2x3, 2-stream system and 2x2, 2-stream system. These test results indicate the same average 2x3-system improvement as the simulations.

Click here for Figure 4

The results were further substantiated by comparing the two systems in a real-world environment (Figure 5). Again, this test shows the same advantage for the 2x3 system.

Click here for Figure 5

3x3 throughput
The performance of the 3x3 device profiled here surpasses any existing standard or proprietary WLAN solutions. A single-band system with traditional 20-MHz channels can use the device to achieve 150-Mbps raw data rate for about 75-Mbps real end-user throughput. When using the device’s 40-MHz-channel mode that will likely be an option under 802.11n, systems get 300-Mbps raw data rate for ~150-Mbps TCP/IP throughput (Figure 6).

Click here for Figure 6

These data rates and the 44 percent average performance improvement tell only part of the 3x3 MIMO story. Equally important is an improvement in the dispersion of throughput performance. Depending on the physical environment, a 2-receive-chain system may not get two uncorrelated signals. In this case, a 2x2, 2-stream system is reduced to a single-stream system. In contrast, the third antenna in a 3-receive-chain system has a much higher probability of getting two uncorrelated signals in the same environment, thus enabling the use of 2-stream mode.

The bottom line is that 3x3 MIMO offers a consistent, high-throughput connection throughout and well beyond the home. One way to think about the performance results across the range of a typical home is that users can expect about 50 percent better throughput anywhere in the home.

The 3x3 performance/cost/power sweet spot
The performance of MIMO systems clearly improves with additional diversity (i.e., more transmitters and receivers). But are these improvements cost-effective? Figure 7 compares system cost with performance and shows that a 2-stream MIMO system’s performance and cost track almost linearly up to a 3x3 configuration. Beyond that point, however, the cost grows considerably without contributing much to performance.

Click here for Figure 7

The cost in terms of both dollars and power dissipation are closely correlated because each additional transmitter or receiver has both a dollar and a power cost. Each transmit chain requires an antenna, power amplifier (PA), RF mixer, RF filter, IF mixer, IF filter, baseband filter, DAC, and some incremental digital circuitry. Each receive chain requires an antenna, low-noise amplifier (LNA), RF filter, RF mixer, IF filter, IF mixer, baseband filter, ADC, and digital circuitry to process the spatially multiplexed signals.

The total cost (dollar amount and power) is roughly a linear function of the number of transmit chains, the number of receive chains, and the number of data streams as well as the square of the number of data streams. The last of these factors must be included because the computational complexity of extracting data streams increases according to the square of the number of streams. In formal terms, cost is proportional to f_c(N, M, S, S²), where f_c is a linear combination function.

While the cost increases roughly linearly with the addition of transmit/receive chains, performance does not. The performance boost contributed by each additional transmitter or receiver tends to diminish past a certain number. One reason for these diminishing returns is that spatial multiplexing only works when the signal streams are statistically uncorrelated. Analysis shows that only so many uncorrelated signal paths are available to a real-world device, because the closer the spacing between antennas, the more correlated the signals get.

At the 2.4-GHz carrier frequency, for example, the wavelength of the radio signals is about 12 cm. Typical APs are 16-20 cm wide. If the separation between each receive antenna is less than 6 cm (half a wavelength), the signals at those antennas will likely have a large degree of correlation, which limits the benefits of spatial multiplexing.

Conclusion
In summary, simply adding more antennas and receivers does not improve the performance linearly, but rather saturates up to the number of uncorrelated signals collected. Therefore, performance is proportional to f_p(S, I x [1-e^-(N+M)]), where S = number of streams (uncorrelated signals measured by the receive chains), N = number of Tx chains, and M = number of Rx chains.

As Figure 7 shows, a 3x3, 2-stream system represents an ideal configuration for the typical WLAN. By integrating the additional transmit and receive chains on a single chip, the incremental cost is kept low, yet performance is considerably enhanced. In fact, a cost analysis model estimates that the bill of materials for a 3x3 AP can be about the same as a conventional 2x2 device for reasons involving the WLAN chip’s level of integration and other factors.

For 2-stream spatially multiplexed MIMO systems, both the 2x3 and 3x3 architectures offer significant performance benefits over the conventional 2x2 configuration for little additional cost. Thus, for access points with larger power budgets, 3x3 is the optimal configuration, while client solutions benefit the most with a 2x3 configuration.

Winston Sun has worked in the wireless telecom/datacom industry for over 10 years as both a systems engineer and algorithm developer. He's worked on CDMA handsets and Wi-Fi as well as GSM base station modem chipsets. He earned his BSEE from UCLA, and MSEE and PhD EE from University of CA, Berkeley. He can be reached at wsun@atheros.com.