Tsoulos George (ред.) MIMO System Technology for Wireless Communications

Подождите немного. Документ загружается.

294 MIMO System Technology for Wireless Communications

a channel makes a trajectory in the complex plane that goes near the origin.

If the pilot symbols are distorted by ISI, the reconstructed channel trajectory

will be in error. When the channel gain is relatively large, a small bias in

channel estimates will not have a big impact. When channel gain is near

zero in amplitude, a small bias in the channel value can actually result in a

deterministic error. This is observed on the left side of Figure 10.14 where

FIGURE 10.13

An example of an error ﬂoor in a space–time modem. (a) Frame error rate, (b) Scatter plot.

−1

−2

−3

0 5 10

Eb/NO dB

15 20

8000

6000

4000

2000

−2000

−4000

−6000

−8000

−6000

−4000

−2000

2000

4000

6000

8000

(b)

(a)

Benchtest f

= 20 Hz

Benchtest f

= 32 Hz

Simulation f

= 20 Hz

Simulation f

= 32 Hz

FER

4190_book.fm Page 294 Tuesday, February 21, 2006 9:14 AM

Multi-Antenna Testbeds for Wireless Communications 295

the actual channel estimation error is relatively small in magnitude but will

cause a very high error rate since the phase channel estimate is roughly 180°

out of phase with the true channel estimate. Simulation showed that this

bias in channel estimation leads to an error ﬂoor. Without companion sim-

ulation and emulation systems, this error ﬂoor would have been tough to

diagnose.

10.4.6 Frequency Offset Calibration

Accurate frequency offset calibration (i.e., compensation) is fundamental in

channel sounding applications of testbeds. In these cases, the transmissions

rarely have provisions for running synchronization algorithms at the receiver.

Therefore, the frequency offset must be measured between all transmitter-

receiver branch pairs prior to the ﬁeld measurement campaign. This section

describes the corresponding procedure.

The sequel derives the mathematical model for the effect of frequency

offset among transceiver branches in the most general case, which corre-

sponds to the second approach discussed in Section 10.4.1 on distribution

of signals in MIMO systems. The ﬁrst and third approaches are special cases

of the results that follow.

The baseband signal at the m

transmitter branch is modeled as:

(10.3)

where is the relative frequency and relative initial phase

modulation introduced by the transmitter branch m with respect to the ideal

carrier frequency and phase. x

(t) is the nominal transmit signal. Similarly,

the baseband signal at each receiver branch is modeled as:

FIGURE 10.14

Causes of performance degradation in rapid fading.

Ci(t)

True channel

Channel due to ISI

Pilot position

Channel estimate

Cq(t)

C(t)

xktxt

mm m



() ()

kt e

()=

+2U^ V

4190_book.fm Page 295 Tuesday, February 21, 2006 9:14 AM

296 MIMO System Technology for Wireless Communications

(10.4)

where is the relative frequency and relative initial phase mod-

ulation introduced by the receiver branch n and y

(t) is the nominal received

signal. The frequency offsets and phase offsets are represented by the vari-

ables {^

, ^

} and {V

, V

}. The time index, t, is implied in the variables {k

, g

}

for notational simplicity.

The frequency and initial phase offset are incorporated in the MIMO signal

models according to:

(10.5)

where the distorted channel matrix is given by:

(10.6)

which assumes the channel matrix H to be static during the observation time,

and with K and G corresponding to the unitary (time-dependent) distortion

matrices due to the frequency offsets at transmitter and receiver branches,

respectively. K and G are given by:

(10.7)

The model of Equation 10.6 shows that all the frequency offset distortion

can be removed at the receiver. This requires estimating all the )

(m, n) =

– ^

and all the )

(m, n) = V

– V

. For channel capacity calculations, the

initial phase offset between the m

transmit branch and n

receiver branch,

)

(m, n) = V

– V

, can be neglected. Note that for an M × N MIMO system,

there are MN frequency offsets between the transmitter and receiver radios

that must be determined.

ygtyt

nn n



() ()

gt e

()=

+2U^ V

YYt

HtXt Nt

HX N









()

() ()

()









HHt

KHG

kh g kh g kh g

kh g k

()

1111 112 2 11

2211 22222

11 1

khg kh g

MMMNN











4190_book.fm Page 296 Tuesday, February 21, 2006 9:14 AM

Multi-Antenna Testbeds for Wireless Communications 297

A method for estimating all )

(m, n) and )

(m, n) in MIMO-OFDM sys-

tems by measuring Inter-Carrier Interference (ICI) is described next. The

effect of a frequency offset in OFDM systems is to cause ICI at the receiver’s

FFT output. The sensitivity of the ICI to frequency offset is illustrated in

Figure 10.15, where the ICI power is plotted for a single subcarrier as a

function of a parameter called the tuning offset, )

(m, n). The tuning offset

sweeps all the possible compensation values for )

(m, n). When it takes the

value )

(m, n) = –)

(m, n), it eliminates ICI, revealing )

(m, n).

With this idea, benchtop calibration of the MN frequency offsets in a

MIMO-OFDM system is attained using a pilot signaling scheme similar to

that described in [33]. This scheme uses pilot tones distributed orthogonally

in space and time. The transmitter branches broadcast these pilot signals

over the air, and the received signal in the n

receiver is tuned across a range

of frequencies as shown in Figure 10.15. The ISI due to the m

transmitter

will be minimized when )

(m, n) = –)

(m, n). Repeating this tuning proce-

dure for each of the MN transmitter/receiver pairs will then fully character-

ize the frequency offset between all radios. Finally, the frequency offset may

vary slowly as a function of time due to changes in temperature in the radio

hardware. In this case, the characterization must be repeated regularly.

10.4.7 Frequency Response Calibration

Wideband transceivers almost never have ﬂat frequency responses in their

passband. Figure 10.16 shows the measured frequency response for one of

the upconverter radios used in the 2.4 GHz MIMO-OFDM testbed. There is

FIGURE 10.15

Power as a function of tuning frequency for a single subcarrier in an OFDM system ()

(1,1) =

12.5 kHz).

−50 −40 −30 −20 −10 0 10 20 30 40 50

−80

−75

−70

−65

−60

−55

−50

−4

Tunin

oﬀset (kHz)

ISI power(dB)

4190_book.fm Page 297 Tuesday, February 21, 2006 9:14 AM

298 MIMO System Technology for Wireless Communications

a notable 2-dB slope and ripples across the passband. More severe cases with

a 5-dB change across the passband were measured for other radio units. The

ripple and slope in the transceiver’s power spectrum is not usually an

important issue for tests involving error rate measurements in data commu-

nications because these non-idealities are considered part of the channel

response. However, uncharacterized non-ideal power spectra are problem-

atic when the testbed is used for MIMO channel sounding. For such appli-

cations, each radio’s frequency response must be characterized as part of the

calibration procedure. The basic procedure to calibrating a M × N MIMO

transceiver is outlined next.

10.4.7.1 Calibration of Transmitters

The strategy is to generate a known signal that can be used to probe the

transceiver’s spectrum across the entire transmission band. Two approaches

are feasible for generating such a signal: a sweep-frequency tone generator

or a wideband signal generator together with a spectrum analyzer.

The swept-frequency tone generator approach is simplest and involves

exciting the transmitter with a single tone whose frequency is stepped at

regular frequency intervals across the input band. An attenuator is placed

between the transmitter and spectrum analyzer to reduce the signal power

to a level that is safe for the analyzer. The spectrum analyzer then records

the (attenuated) transmitter output signal in peak-hold mode. Figure 10.16

is an example of the resulting spectrum using this technique. The disadvan-

tage of this method is that the peak-hold function records the larger peak

noise, instead of an average noise.

The second approach is to excite the transmitter with a set of evenly spaced

tones across its input band. This approach requires a wideband signal gen-

erator that can synthesize a user-deﬁned signal. A suitable signal design is

proposed in [33]. The spectrum analyzer then records the transmitter’s out-

put in averaging mode. The beneﬁt of this method is that zero mean AWGN

is cancelled out by averaging. The drawback is that the equipment required

to generate the wideband excitation signal can be more sophisticated and,

FIGURE 10.16

The measured frequency response spectrum of a transceiver radio in (a) transmit mode and

(b) receive mode.

−12 dBm −27 dBm

2.43 GHz

2.45 GHz

(a) (b)

8 dBm

−12 dBm

60 MHz

80 MHz

4190_book.fm Page 298 Tuesday, February 21, 2006 9:14 AM

Multi-Antenna Testbeds for Wireless Communications 299

hence, more expensive than the signal generator used to generate a swept-

frequency tone. Another beneﬁt of using the wideband signal generator in

[33] is that the multiple transmitters in a MIMO system can be calibrated

simultaneously using a M-to-1 RF combiner and a single spectrum analyzer.

This is possible because spatial orthogonality in the excitation signal allows

the combined TX spectra to be separated in post-processing. An example of

the spectrum generated by this approach for a case with four radios is shown

in Figure 10.17. The four transmitter radios used are the same design as the

one shown in Figure 10.16a.

10.4.7.2 Calibration of Receivers

The procedure for calibrating the receivers is the same as that for calibrating

the transmitters except that connections to the inputs and outputs are

swapped. In the case that the signal generator cannot generate an excitation

signal at the radio frequency required by the receiver radio’s input, a calibrated

transmitter radio can be used to up-convert the excitation signal. The trans-

mitter’s output can then be connected through an attenuator to the receiver’s

input.

Once the frequency responses of all transmitter and all receiver branches

have been quantiﬁed, precise channel sounding measurements can be con-

ducted by pre-compensating the signal transmitted on each transmitter

branch, and by post-compensating the received signal of each receiver branch.

10.5 Field Test Administration

Once a testbed is fully calibrated in the laboratory and for its intended

environment, it is ready to be deployed for ﬁeld tests and measurements.

FIGURE 10.17

(a) Measured power spectrum at the transmitter antenna port before pre-equalization by the

transmit DSP and (b) after pre-equalization.

−21

TX1

TX2

TX1

TX2

−22

−23

−24

Power spectrum(dBm)

−25

−26

−21

−22

−23

−24

Power spectrum(dBm)

−25

−26

2.43 2.45

(a)

(b)

Frequency(GHz)

2.43 2.45

Frequency(GHz)

4190_book.fm Page 299 Tuesday, February 21, 2006 9:14 AM

300 MIMO System Technology for Wireless Communications

At this point, the logistics of testbed deployment and data collection become

an important issue. A well-designed Graphical User Interface (GUI) greatly

facilitates setting up the testbed in the ﬁeld and monitoring its operation.

Similarly, the collection of data tends to be cumbersome and labor-intensive

when performed manually, while automating the process allows for exten-

sive data collection in a variety of environments in an efﬁcient and reliable

way. Both topics are addressed in the following sections.

10.5.1 Automated Field Measurements

The following example illustrates the value of automating measurements in

an experiment that tested space–time codec performance. Consider the chan-

nel experienced by a receiver as its antenna array moves within a square

area of 10 wavelengths on a side. The channel variation in such a small area

is due mainly to multipath interference. A typical spatial sampling pattern

used for testing the aforementioned space–time codec is speciﬁed below in

Table 10.3. For comparison, equivalent requirements for testing a SISO codec

are provided. For each spatial sample within the area, a full range of SNRs

is required to stress the codec in both the low and high capacity regions of

the channel realization associated with that spatial sample. This sampling

strategy ensures that the codec is subjected to a sufﬁcient range of channels

and noise realizations to give a good approximation of its performance at

the microscopic (local) level. Several areas have to be measured in this way

to subject the codec to a range of macroscopic channel characteristics in the

larger environment.

A measurement proceeds according to the pseudo code given below. Lines

2–10 in the Measurement Pseudo Code can be executed automatically by a

single component in a MIMO testbed provided that (1) all components of

the testbed are able to communicate and (2) the spatial sampling can be

performed robotically. The 5 seconds allocated to each measurement (cf.

Table 10.3) include the time to execute lines 3–6 in the pseudo code. It is

clear from this example that acquiring a large data set requires a signiﬁcant

TABLE 10.3

Measurement Budget

SISO Case MIMO Case

Sampling grid size 10Q x 10Q square 10Q x 10Q square

Spacing between grid points 0.5Q 0.5Q

Antenna array geometry Not applicable Linear

Antenna spacings Not applicable 0.5Q, 1Q, 1.5Q, 2Q

Total number of spatial samples 21*21 = 441 21*21*4 = 1764

Measurement interval 5 seconds 5 seconds

Measurement time per grid 441*5 = 38 minutes 1764*5 = 2.4 hours

Number of locations in environment 10 10

Total Experiment Time 38*10 = 6 hours 2.4*10 = 24 hours

4190_book.fm Page 300 Tuesday, February 21, 2006 9:14 AM

Multi-Antenna Testbeds for Wireless Communications 301

time investment. It would be impossible to manually place the antenna array

and to trigger the radio hardware with the allotted time per measurement

(and, at any rate, it would be a very inefﬁcient use of qualiﬁed personnel).

Measurement repeatability and system reliability are other essential charac-

teristics that would be challenging for the test engineer to achieve but are

easily achieved with automation.

A testbed conﬁguration with the above capabilities is illustrated in Figure

10.18. Robotic antenna arrays are responsible for spatial sampling by moving

the transmitter and receiver antenna arrays according to the spatial pattern

deﬁned in Table 10.3. TCP/IP connectivity allows any terminal on the Inter-

net to control both radios and robots remotely.

An effective and affordable solution to automating spatial sampling can

be realized from LEGO

Mindstorms components and open source software

[32]. Two such robots are shown in Figure 10.19 and Figure 10.20. These

robots have a 1.59 mm positioning resolution and a 70 × 70 cm positioning

range. These dimensions are well suited to communications within the

1–10 GHz band where wavelengths (30–3 cm) are on the same order of

magnitude that can be achieved by Lego structures. The Lego system allows

the robots to be conﬁgured modularly so that as many as four antennas can

be move independently in azimuth. The software that controls the robots is

comprised entirely of Java-based code and allows the end user to control

the robot with high level commands (move, reset, calibrate, etc.). These

commands can be issued via TCP/IP as text strings, thereby enabling control

from any application that supports TCP socket communications (C++,

MATLAB, web browser, etc.).

Remote control of the robots is enhanced by using a web camera to stream

live video of the robot’s activity over the Internet. In the case of a mechanical

failure, the robots can alert the remote user of the problem by e-mail. This

affords the user the option of checking the experiment’s progress intermittently,

Measurement Pseudo Code

1 For each location in the environment

2 For each antenna separation

3 For each point in the sample grid

4 Encode data and transmit signal

5 Receive signal decode data

6 Tabulate information error rate

7 Move to the next point in sample grid

8 end

9 Change array geometry

10 end

11 move to the next location in the environment

4190_book.fm Page 301 Tuesday, February 21, 2006 9:14 AM

302 MIMO System Technology for Wireless Communications

instead of requiring constant vigil over long and tedious experiments.

Another great convenience of the automation system is that it allows a single

user to simultaneously control and monitor both the transmitter and receiver

FIGURE 10.18

A highly automated MIMO testbed.

FIGURE 10.19

Four-antenna transmitter and robotic antenna array.

Space–time

digital radio

Transmitter node Receiver node

Space–time

digital radio

Video

camera

Video

camera

Control

terminal

Control

terminal

Robotic

antenna array

Robotic

antenna array

Radio

channel

TCP/IP internet

Remote

control

terminal

Operator

(anywhere)

4190_book.fm Page 302 Tuesday, February 21, 2006 9:14 AM

Multi-Antenna Testbeds for Wireless Communications 303

units even though they may be located in different parts of a building or

neighborhood. This would otherwise require a minimum of two users.

10.5.2 Graphical User Interface Design

One vital feature of a testbed is to have well-designed Graphical User Inter-

face (GUI) for controlling its operation. In the authors’ experience, the test-

bed’s capabilities can be exploited at a maximum if the GUI provides the

following features:

1. Immediate (quasi real-time) feedback of test results and environ-

mental conditions

2. High ﬂexibility for conﬁguring and controlling system parameters

“on the ﬂy”

3. A clear and easy-to-understand interface for demonstrating the test-

bed’s capabilities and features to third parties

Each of these three points is described in detail next.

Quasi real-time feedback: A GUI’s capability for reporting transmission

results and testbed status in real-time or immediately following transmission

simpliﬁes the data collection and speeds up the measurement process. It

maximizes the efﬁciency of ﬁeld measurement campaigns by allowing for

early detection of system failure, quick troubleshooting, and on-site assess-

ment of the environment’s validity for the desired test (e.g., channel condi-

tion, SNR, etc.). Useful information to display includes error statistics, signal

and noise power, scatter plots, instantaneous channel state information,

channel capacity, etc.

FIGURE 10.20

Four-antenna receiver radio and robotic antenna array.

4190_book.fm Page 303 Tuesday, February 21, 2006 9:14 AM