Multipath expands RF bandwidth

At present, third-generation (3G) mobile wireless networks are designed to offer up to 2-Mbit/second data rates. Multiple-input, multiple-output (MIMO) technologies, however, can boost those data rates to 14.4 Mbits/s and potentially even higher. This is achieved by utilizing the available spectrum in a more-efficient manner with an array of antennas at both transmitter and receiver.

Spectrum is a limited resource in any wireless system. Recently there has been a considerable amount of work done using multiple antennas to more efficiently utilize the available spectrum. One of the simplest methods of achieving this is to use selection diversity at the receiver, where the receiver selects the antenna that has the highest receiver sensitivity. On the transmit side, multiple antennas can be used for space-time coding for transmit diversity. This provides additional margin in the communication link budget and can reduce the transmit power, increase the system range or increase network capacity.

A further step is to utilize intelligent antennas (IAs) for beam forming to increase system capacity significantly by reducing the level of interference between users. IA is already being deployed into mobile wireless-infrastructure products. Multiple-antenna communication systems are here to stay, and as the cost of integrated electronics for RF and baseband processing decreases, multiple antennas will quickly become mainstream in wireless systems.

While multiple antennas used in wireless systems to date have had an impact on system capacity and receiver sensitivity, there is a significant opportunity for wireless communications to increase throughput over a given channel.

MIMO systems use an array of transmit and receive antennas for enormous gains in spectral efficiency by exploiting a rich multipath fading environment. The systems split a single user's data stream into multiple substreams and use an array of transmit antennas to simultaneously transmit the streams into the same frequency band. While the signals would interfere with one another in a single-antenna system, it is possible to use the scattering of those signals to enhance, rather than degrade, transmission accuracy. The scattered paths use spatial signatures to create separate, parallel subchannels.

At the receiver, an array of antennas detects the multiple transmitted substreams. Using the MIMO technique, the rate of transmission is increased in proportion to the number of antennas used to transmit the signal. Furthermore, previous techniques to orthogonalize channels, like code-division multiple access (CDMA), can be laid on top of MIMO systems to ensure the bandwidth can still be a shared resource. A MIMO system can be added to a 3G system in a seamless manner, boosting the data-carrying capacity of the network without impacting the other 3G services.

Targeting wireless standards
To that end, CDMA-MIMO is currently being considered for both the 3GPP and 3GPP2 mobile wireless standards. The High Speed Downlink Packet Access (HSDPA) standard for the Universal Mobile Telecommunications System uses the same 2.1-GHz spectrum as regular UMTS but allows the use of higher constellations and shorting spreading codes to achieve high data rates to the portable device. A working group is currently proposing a MIMO-HSDPA standard.

One of the major factors holding back the standardization of MIMO systems is the complexity and performance of the receiver design. The computational complexity of a MIMO receiver increases dramatically with the number of antennas in the system and the type of constellation used. In an asymmetric standard like HSDPA, the receiver must be embedded into a portable device like a PDA or PCMCIA card. The cost and power consumption of the receiver is therefore critical to enable widespread commercial deployment. It all comes down to the type of detection algorithm used.

The earliest detection algorithm for MIMO signals was the Vertical Bell Labs Layered Space-Time (VBLAST) algorithm. The algorithm was intended for very high antenna configurations, i.e., 16 x 16. At its core, VBLAST is an iterative cancellation method that depends on computing a matrix inverse to solve the zero-forcing function. The algorithm works by detecting the strongest data stream, subtracting that stream from the received signal and repeating the process for the remaining data streams. While the algorithm complexity is linear with the number of transmit antennas, it suffers performance degradation through the cancellation process. If the cancellation is not perfect, it can inject more noise into the system and degrade detection.

Other proposals for MIMO detection include using an iterative process around a linear receiver, much like the way in which turbo decoding works to converge to a solution. The iterative method with coding helps to refine the cancellation process and ensure that the correct data stream is canceled.

While iterative detection can increase receiver sensitivity, there are substantial problems with a real implementation. The iterative loop must pass through the physical and transport channel block interleaving that exists between the demodulation and channel decoding. Furthermore, soft information from the decoder that is to be sent back to the detector must first be re-interleaved. This requires substantial buffering of baseband samples and introduces an unacceptable latency in the detection process.

If you consider today's 3G wireless communications system, it is more reasonable to expect MIMO systems with small numbers of antennas. For example, the MIMO-HSDPA working group in the 3GPP standard is considering up to four antennas. A mobile PDA can easily fit two antennas in its form factor, and a notebook screen can support four antennas integrated into the case. Instead of pursuing suboptimal linear receivers or complex iterative schemes, it is better to leverage technology advances in silicon to use the optimal detection strategy for MIMO.

The optimal detection strategy for a MIMO receiver is to perform a maximum-likelihood (ML) search over all possible transmitted symbol sets. To date, such an approach has been considered too complex to implement for high data rates.

The complexity of the search is related to the modulation format and the number of independent transmit antennas, where the number of search states is 2MQ-M is the number of transmit antennas and Q is the bits per symbol. For a 4 x 4 quadrature phase-shift keying (QPSK) system, there are 256 possible sets of unique bits that can be transmitted per symbol period. But, in channel simulations on a 4 x 4 QPSK system, the ML search can provide more than a 5-dB gain over the VBLAST algorithm for uncoded transmissions.

Testing silicon
At our research facility in Australia, we have tested a silicon device that performs ML detection and produces soft outputs approximating the log likelihood ratios (LLRs) of the posteriori probabilities (APPs) for each of the transmitted bits. This algorithm is termed ML-APP. The ML search is achieved with a parallel-processing architecture that exploits the symmetry of the system.

The core of the MIMO detector chip is only 7.3 mm2 in 0.18-micron CMOS. It consumes 190 milliwatts when performing ML-APP detection for a 4 x 4 QPSK MIMO channel at a data rate of 19.2 Mbits/s. At that data rate, a bit-error rate (BER) of 10-3 has been achieved with a ratio of energy per bit to the spectral noise density (Eb/N0) of 2.5 dB.

Our investigations into MIMO communications systems have produced some interesting and unexpected results. There are several MIMO configurations that are capable of sending 8 bits over the air simultaneously. Two examples are four transmit antennas each with QPSK, or two transmit antennas with 16 -QAM. In simulations with a fading environment, the 4 x 4 QPSK system showed superior signal-to-noise ratio (SNR) performance as compared with a 2 x 2 16-QAM system.

We concluded that where there is a rich scattering-channel environment, the receiver can maintain a fixed throughput in a much lower SNR environment by utilizing additional transmit and receive antennas instead of using higher high-constellation formats. Furthermore, experiments with a receiver containing only two receive antennas show that using 4 x 2 QPSK outperforms 2 x 2 16-QAM even though the same data rate is achieved.

We will eventually see portable data devices deployed with two or four antennas. When that happens MIMO can be used to substantially increase the throughput and achieve even higher data rates.

See related chart