Traditionally, multipath interference was regarded as the nemesis of non-line-of-sight communications, due to pronounced fading phenomena. However, recent studies have demonstrated that capacity of channels possessing a rich multipath structure may be exploited with multiple antennas on both sides of the communications link. Multiple antennas gives rise to the field known as space-time channel coding, coding in time as well as space to take advantage of multipath interference.
This page presents a brief explanation of the narrowband MIMO measurement system developed at BYU. After explaining the need for channel measurements, we present an overview of the hardware and describe the transmit and receive systems in detail. A number of data sets have been collected with the platform, leading to important discoveries about multiple-antenna systems. Further details are available in our Publications.
Initial studies in space-time coding relied heavily on very simple analytical models to assess channel capacity and coding performance. However, these models may be too simplistic to capture the behavior of real-world MIMO channels. We propose that direct measurement of a variety of channels is required for at least two reasons: the complexity of real-world channels and the need to validate simpler models.
The ultimate goal of wireless propagation research is the ability to model all important behavior for any particular environment. Numerical studies involving ray tracing have demonstrated the ability to match bulk propagation behavior for a number of important MIMO channels. However, determining more sensitive channel behavior such as phase correlation and channel stationarity is difficult with such techniques.
Direct measurement of MIMO channels serves as a solid source of ground truth for assessing new MIMO channel models.
A schematic diagram of the platform is shown below.
Each transmit antenna is excited by a high-frequency carrier. The carrier for each antenna is modulated by a separate PRN sequence. On the receive side, the signal of each receive antenna is mixed down to an appropriate IF, sampled, and stored on a computer. Channel estimation based on the data is then performed in software off-line.
The transmitter consists of a digital pattern generator, a microwave source, and a box containing custom-built RF components.
The digital pattern generator and microwave source both feed into
the custom RF chassis. Inside the chassis the signal from the microwave
source is split into 16 channels, amplified, and modulated by the digital
patterns. Each of the resulting signals is fed to the antenna array.
We have performed measurements with two types of antennas. Our initial measurements were carried out with a dual-polarization patch array tuned to 2.45-GHz, with half-wavelength inter-element spacing. Each antenna is matched with a 3/4 wavelength section which acts as a band-pass filter on the receive side. We have also performed measurements at 2.43-GHz with monopole antennas that are positioned in a 33x33 grid. Quick reconfiguration of the antennas in the grid allows many different antenna array to be realized.
The transmit chassis is fed by the microwave source (LO input) and the digital pattern generator. The LO input feeds a backplane where the signal is divided equally among the 16 channels. A number of transmit cards are plugged into the backplane which perform mixing and signal and power amplification. Finally, each of the signals is fed to a transmit antenna.
Both the transmit chassis and the receive chassis contain an RF backplane. The backplane employs a resistive power divider to obtain broadband power division up to 10 GHz. Each card slot is fed by a surface mount SMA connector and an edge card connector supplying power and a baseband or IF signal.
The transmit card depicted above amplifies the LO and mixes with the baseband signal (+V/-V in this case). The resulting signal is fed through a preamp and a partially matched power FET. The card has been designed for broadband operation from 0-3GHz.
The receiver consists of many of the same basic components as the
transmitter. Signals from the antenna array are fed into the RF
chassis where they are mixed down to an IF by individual receiver
cards. The resulting signals are sampled by an A/D board inside
of a PC and stored on disk.
The receive chassis has a similar layout the the transmit chassis. The LO input is fed into the backplane. Signals from the receive antennas are fed into receive cards and mixed down to an IF. The IF signals are fed (via the backplane) into an IF board which provides digitally-controlled gain as well as low-pass filtering. The resulting signals are fed via the A/D connector to the PC.
The IF board accepts the 16 possible IF inputs (only 9 channels are populated in the picture. Each signal is filtered by simple RC networks and amplified by digitally-controlled variable gain amplifiers.
The receive card amplifies by the signal from the antenna and mixes down to an IF. The resulting signal is fed via the edge card connector to the IF board.
A number of data sets collected with this platform are available for download.
This subsection outlines a number of important findings on MIMO channels. For further details, consult the publications listed below.
Very simplistic models assume that capacity grows linearly as the number of antenna elements is increased at transmit and receive. The following figure depicts the complimentary cdfs (CCDFs) of capacity per number of antennas as more elements are packed into an array constrained to be 2.25 wavelengths long for 20 dB SNR, compared with the standard i.i.d. Gaussian model.
Evidently, as we pack more antennas into the fixed aperture, adjacent antennas become more and more correlated, and the linear capacity increase of the ideal model is not attained.
Arrays constructed with dual-polarization elements may be able to accommodate more antenna elements in the same aperture than single-polarization arrays. Also, since depolarization is small for structures built mostly with right-angle geometries, the two orthogonal polarizations will be nearly independent in a deterministic sense. However, orthogonal channels also present the possibility of reduced receiver power.
Measurements were taken for two dual-polarized elements separated by one-half wavelength at transmit and receive. The following figure plots capacity CCDFs for measured channels with various types of polarization: single polarization (SP), dual-polarization without separation (DP), and dual-polarization with separation (DPS). Also, the results of Monte Carlo simulations for two ideal channels are presented for comparison: the i.i.d. orthogonal dual-polarization channel (DP) and the i.i.d. single polarization channel (SP).
Obviously SP is the clear winner, since we have equal average power transfer from each transmit to receive element and independent fading. Although the DP model provides orthogonal (diagonal) channels, power is cut in half at the receiver, leading to an overall capacity degradation. Real-world channels tend to fall somewhere in between. Measured single-polarization channels (SP) exhibit some correlation between antenna elements, leading to capacities less than (SP). The measured dual-polarization channel (DP) suffers some power loss due to a more diagonal structure, but the lower correlation offsets this effect, leading to a net gain in capacity.
To assess the impact of antenna directivity, we compared the capacity CCDFs for a 4-element monopole (omnidirectional) array with a 4-element patch (directional) array where an SNR of 20 dB has been assumed.
Interestingly, the two systems have almost the exact same capacity. Therefore both types of antennas "see" about the same amount of multipath, meaning that the multipath is fairly rich. We realize however, that forcing both systems to have the same SNR of 20 dB may be unwarranted, since proper alignment of the patch antennas could yield a capacity improvement, as suggested by the curve to the far right.
MIMO studies have focused on how rich multipath leads to high capacity. However, this is only necessarily true for a fixed SNR. Multipath results from increased scattering, which would tend to lower the receiver SNR. Therefore, one might expect SNR and multipath richness to be competing goals. The following figure plots a map of the capacity for different channels probed from many different locations in a building. Each arrow indicates a 10x10 channel measurement from one point to another. The box on each arrow has two different capacity vales. For the top number, each channel has been normalized to attain an SNR of 20 dB, effectively negating any differences in path loss. For the bottom number, all channels have been normalized by the same number, so that the average SNR over all channels is 20 dB, thus accounting for path loss. Quite different conclusions would be drawn from the two numbers about what constitutes a "good" channel. Generally, we see an inverse relationship between the SNR and the amount of multipath. However, often when the transmitter and receiver are in close proximity, the multipath is still quite large.
| Date | Title | Source | Authors | Download/Info |
|---|---|---|---|---|
| 06/03 | On signal strength and multipath richness in multi-input multi-output systems | ICC'03 | T. Svantesson, J. Wallace | |
| 11/02 | Statistical characterization of the indoor MIMO channel based on LOS/NLOS measurements | Asilomar'02 | T. Svantesson, J. Wallace | |
| 07/02 | Experimental Characterization of the MIMO Wireless Channel: Data Acquisition, Analysis, and Modeling | IEEE Transactions on Wireless Communications | J. Wallace, M. Jensen, A. Swindlehurst, B. Jeffs | |
| 05/02 | MIMO wireless multipath ray parameter estimation from channel transfer matrix measurements | ICASSP'02 | B. Jeffs, J. Wallace | |
| 10/01 | Measured Characteristics of the MIMO Wireless Channel | IEEE VTC'2001 | J. Wallace, M. Jensen | |
| 05/01 | Spatial characteristics of the MIMO wireless channel: experimental data acquisition and analysis | IEEE ICASSP'2001 | J. Wallace, M. Jensen | |
| 01/01 | Experimental Characterization of the MIMO Wireless Channel | IEEE Antennas and Propagation Symposium | J. Wallace and M. Jensen | abstract, pdf |
| 01/01 | Characteristics of Measured 4x4 and 10x10 MIMO Wireless Channel Data at 2.4-GHz | IEEE Antennas and Propagation Symposium | J. Wallace and M. Jensen | abstract, pdf |