measurements and models of radio frequency...

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 11 , NO. 7, SEPTEMBER 1993 99 1

Measurements and Models of Radio Frequency Impulsive Noise for Indoor Wireless

Communications Kenneth L. Blackard, Member, IEEE, Theodore S. Rappaport, Senior Member, IEEE, and Charles W. Bostian, Fellow, IEEE

Absfracf- This paper presents the results of average and impulsive noise measurements inside several office buildings and retail stores. The noise measurement system operated at 918 MHz, 2.44 GHz, and 4 GHz with a nominal 40 MHz 3 dB RF bandwidth. Omnidirectional and directional antennas were used to investigate the characteristics and sources of RF noise in indoor channels. Statistical analyses of the measurements are presented in the form of peak amplitude probability distributions, pulse duration distributions, and interarrival time distributions. Simple first-order mathematical models for these statistical characterizations are also presented. These analyses indicate that photocopiers, printers (both line printers and cash register receipt printers), elevators, and microwave ovens are significant sources of impulsive noise in office and retail environments.

I. INTRODUCTION

MPROVEMENTS in RF technology have spawned a de- I mand for inexpensive, easy-to-deploy wireless indoor communication systems and products which require less time to install and usually cost much less than wireline systems.

Researchers have given considerable attention to the investigation and modeling of indoor radio wave propagation in recent years [1]-[4]. However, little scientific work has been done to determine the significance of indoor radio frequency (RF) impulsive noise and its impact on system performance. Noise models for indoor channels and specific noise sources are important for determining irreducible error rates and cod- ing requirements for indoor communications. Furthermore, if particular devices are known to be noise sources, this knowledge can be used to assist in the successful deployment of indoor wireless networks. The work presented in this paper lays the foundation for modeling the effects of impulsive noise on indoor wireless communications.

Much work has been done to characterize thermal noise and impulsive noise in outdoor mobile and portable radio communications. Extensive measurements and analyses of

Manuscript received March 1992; revised November 1992. This research was sponsored by the NCR Corp. and the MPRG Industrial Affiliates Program. This paper was presented in part at the 1991 ICC, May 1991, Denver, CO.

K. L. Blackard is with the Federal Bureau of Investigation, Quantico, VA 22135.

T. S. Rappaport and C. W. Bostian are with the Mobile Portable Radio Research Group, Bradley Department of Electrical Engineering, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061.

IEEE Log Number 9210924.

automobile ignition noise at HF, VHF, and UHF are well documented in [SI-[lo]. These experiments were made to investigate the effects of impulsive noise on narrowband communications. Hence, the measurements were made using noise receiver bandwidths of 3-20 kHz. We are not aware of other publications which describe impulsive noise measurements inside buildings at UHF and microwave frequencies, using wide bandwidth receivers.

Future indoor wireless communications will most likely operate in the low microwave bands due to the wide bandwidths available there and the existing spectral congestion and higher ambient noise at UHF and below. This paper focuses on three frequency bands likely to be used in future indoor wireless systems: 918 MHz, 2.44 GHz, and 4 GHz. Two of these bands, 918 MHz and 2.44 GHz, lie in the U S . industrial, scientific, and medical (ISM) bands (902-928 MHz and 2.40-2.483 GHz). The ISM bands are already being used for indoor wireless systems, as the Federal Communications Commission (FCC) has allocated these bands as license-free if spread spectrum with less than 1 W of power are used. The third band studied, 4 GHz, lies in C band which is used today for satellite and terrestrial communications.

Impulsive noise occurs in short bursts. In order to produce characterizations that are applicable to all possible future systems, the measurement system must have a very wide bandwidth. Besides being very expensive, a wideband measurement system will experience coherent interference within its bandwidth from unwanted signals in the congested radio frequency spectrum [6]. Nevertheless, the noise measurement system must have, as a minium, a bandwidth at least as wide as that of the proposed communication system. The measurement system used for this paper had a nominal RF bandwidth of 40 MHz, which is greater than the bandwidths of many current indoor wireless systems and on the order of the widest channels likely to be used in the ISM bands.

The objective of this research was to develop empirical radio frequency impulsive noise models based on the results of an extensive measurement campaign. These models can aid in the simulation and design of indoor wireless communication systems and, based on the measured data presented in this paper, have been shown to provide good first-order agreement to actual measured impulse noise waveforms [17]. In this paper, we statistically quantify how impulse noise is impacted

0733-8716/93$03.00 Q 1993 IEEE

992 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 11, NO. 7, SEPTEMBER 1993

uu

A LOCAL OSCILLATOR

W RuQ#QpauToFI

Fig. 1. Block diagram of the three-band noise measurement system.

by operating frequency, building environment, and individual sources.

11. EXPERIMENT DESIGN

A. Noise Measurement System

A three-band noise measurement system was developed and used to characterize average and impulsive RF noise statistics at 918 MHz, 2.44 GHz, and 4 GHz, and to survey these bands for CW or modulated signals. The noise measurement system consisted of a superheterodyne noise receiver, omnidirectional and directional antennas, a spectrum analyzer, a digitizing oscilloscope, and a personal computer. The block diagram of the measurement system is shown in Fig. 1.

The noise receiver incorporated a bank of microstrip bandpass filters, wideband low noise amplifiers, and a logarithmic video detector with a 40 MHz passband centered around 160 MHz which provided approximately 65 dB dynamic range. The 3 dB RF bandwidth of the noise receiver was 40 MHz for the 918 MHz and 4 GHz bands, and 32 MHz for the 2.44 GHz band. The receiver bandwidth was limited by the receiver component with the smallest bandwidth. In the 918 MHz and 4 GHz bands, this was the 3 dB RF bandwidth (40 MHz) of the logarithmic video detector. The 3 dB bandwidths of the 918 MHz and 4 GHz cascaded filter bands were approximately 45 MHz and 65 MHz, respectively. In the 2.44 GHz band, the receiver bandwidth was limited

typical of inexpensive commercial receiver systems and their antenna feeds.

A spectrum analyzer connected in parallel with the noise receiver allowed the system operator to detect CW and modulated signals visually. The oscilloscope digitized the baseband output of the logarithmic detector and stored the digitized waveforms on the computer hard disk.

Ideally, a noise measurement system should acquire data continuously during measurements, much like a strip-chart recorder. However, our noise measurement system was not capable of performing continuous acquisitions because of the memory and timing limitations of the digital oscilloscope. The Tektronix 2432A Digital Oscilloscope required 19 ms between consecutive single-sweep acquisitions to re-arm its trigger circuits. This limited the system acquisition rate to approximately 40 waveforms (512 bins per waveform) per second. Due to our experimental design, this limitation did not reduce the usefulness of our measurements, since three different sweep speeds were used to ensure capture of impulsive noise events with varying durations. This is described in detail in Section 11-D.

Measurements at Sites A-D used a broadband omnidirectional discone antenna with a gain of approximately 1.5 dBi in each band [ l l ] , [12]. In Site E, 12 dBi directional monofilar axial-mode helical antennas were used for each measurement band to find and measure RF characteristics of specific sources as a function of carrier frequency.

by the 32 MHz 3 dB bandwidth of the 2.44 GHz microstrip bandDass filters. The Dassband shapes of the filter cascade in B. Locations and Data

all bands is given in [18]. In all bands, the noise figure of the receiver cascaded with the 2 m length of coaxial antenna feed terminated in a 50 R load was approximately 11 dB, which is

Impulsive noise and in-band CW interference signals were measured inside five different buildings: a large grocery store in Blacksburg, VA (Site A), a major department store in Chris-

993 BLACKARD et al.: MEASUREMENTS AND MODELS OF RADIO FREQUENCY IMPULSIVE NOISE

tiansburg, VA (Site B), two large open-plan soft-partitioned office buildings located in the business district of Dayton, OH (Sites C and D), and Whittemore Hall, a closed-plan hard-partitioned office building on the Virginia Tech campus (Site E).

At Site A, measurement locations were near an operating check-out lane, at the end of a shopping aisle, and near the deli section of the store. At Site B, measurements were made adjacent to an operating check-out counter and at the end of an aisle in the electronics department. At Sites C and D, measurements were made on several different floors of the buildings, near computer terminal rooms, perimeter windows, and in large office areas throughout the buildings. In Site E, measurements were made in a hallway near a photocopier and elevator terminal, and in a student lounge which contained a microwave oven.

In Sites A-D, locations which are representative for future indoor wireless systems were selected as measurement locations. At each measurement location, several measurement runs, each lasting for exactly three minutes, produced thousands of impulsive noise waveforms and millions of impulse noise time bins from one of the three measured bands. Measurement runs for a particular band and location were repeated using antenna heights of 1.75 and 2.25 m above the floor. We used two different antenna heights, which differed by at least a wavelength at the lowest frequency, in order to average out possible frequency-selective fading effects in the received noise waveforms at a particular location.

Measurements conducted at Sites A and B were the most statistically rigorous of the campaign, using three different oscilloscope sweep speeds and two different antenna heights for measurement runs in each of the three measurement bands. A complete set of eighteen measurement runs for all bands and sweep speeds was typically measured within one hour at any location.

At Sites C and D, only one oscilloscope sweep speed was used in order to perform broad noise surveys at a greater number of locations throughout the buildings. In these sites, six measurement runs (consisting of three different frequency band runs with two different antenna heights) constituted a complete set of measurement runs for a particular location. Measurements were made at nearly 20 different locations throughout the two buildings.

At Site E, measurement runs were made in each frequency band over 1 m intervals during the continuous operation of three specific noise sources. The noise sources measured were a pay-per-copy photocopier, an elevator switch, and a microwave oven. Directional helical antennas were used to locate the maximum RF signal from each source. Careful records of the noise source and receiver separation were kept so that propagation models could be developed for each noise source. Statistical results for the entire measurement campaign are given in Section 111.

C. Conducting a Measurement Run

Several times during each measurement day, the noise receiver was calibrated for each frequency band. The calibration allowed signal levels at the receiver input to be

determined from the oscilloscope vertical deflection. A CW signal with known power level and frequency equal to the center frequency of the particular band (918 MHz, 2.44 GHz, or 4 GHz) was applied directly to the receiver’s antenna terminal. The power level of the applied CW signal was varied from -100 to -25 dBm in 5 dB increments, and the average dc signal level at the output of the log detector was measured and recorded on the computer’s hard drive for processing. The system’s bandpass response in each band was also calibrated using a white-noise generator and the receiver spectrum analyzer. The amplitude and bandpass responses in each band were consistent throughout the measurement campaign.

Before each noise measurement run, a 50 R “dummy” load was placed at the receiver antenna terminal. The time-average dc signal out of the log detector was then recorded. The dc level corresponded to the average thermal noise floor of the receiver at that location, and was a function of the noise figure of the receiver. This value was stored for future data processing and varied by less than f l dB throughout the measurement campaign. In addition, the receiver’s thermal noise power waveform was measured and recorded over a three-minute period to determine the statistics of detected thermal noise through our peak detector, and is shown in Fig. 4.

Immediately following the thermal noise calibration, the “dummy” load was replaced with the measurement antenna, and the oscilloscope trigger level was adjusted to a particular level above the new thermal noise floor so that the oscilloscope would not trigger on ambient thermal noise but rather on impulsive noise bursts.

D. Temporal Resolution of Impulsive Noise Peaks

The digitizing oscilloscope used an envelope acquisition mode, which relies on a wideband analog peak detector to measure the peaks of signals within discrete-time intervals. The oscilloscope horizontal sweep rate was set at either 1 psldiv, 100 psldiv, or 10 ms/div, with 20 divisions per waveform. The oscilloscope quantized each swept waveform into 5 12 consecutive time intervals, and the oscilloscope sweep speed determined the duration of each discrete-time interval (bin). Thus, the fastest sweep speed of 1 ps/div provided a discrete-time interval of 20 ps/512, or about 40 ns per bin. If one assumes the minimum measurable pulse duration of a noise burst is equal to 40 ns (which is roughly the reciprocal of the system baseband bandwidth), then these sweep rates correspond to 512, 51 200, and 5 120 000 possible impulse bursts in a single swept waveform. By using a sweep speed which yields a bin duration on the order of the minimum measurable pulse width, the oscilloscope display captured a record of the peak amplitudes of individual noise bursts closely separated in time. Two slower sweep speeds were also used to ensure that noise bursts separated by relatively long times were captured and statistically analyzed.

The following equation relates the time resolution of each bin to the horizontal sweep rate in our measurements.

sweep speed [ns/div] x 20 [div] 512 [bins] time resolution [ns/bin] =

(1)

994 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 11, NO. I , SEPTEMBER 1993

918 HZ 40 nwbin resolvable impulses could have occurred within each 400 ps time bin, even though only one peak value was retained. Therefore, the measured data provide a slight upper bound of the actual impulse duration when in the fast sweep mode, and a very coarse estimate when slower sweep are used.

Analysis of the data from all measurement locations indicate that the most significant sources of impulsive noise in office and retail environments are microwave ovens, photocopiers, printers (cash register receipt printers and line-feed printers), elevator door switches, and gas-powered engines with spark- gap ignition systems.

Impulsive noise was prevalent throughout the grocery store environment at Site A. Sources of the impulsive noise were cash registers, microwave ovens, a gas-powered floor cleaner, refrigeration compressor motors, and line printers.

At Site B, sources of significant impulsive noise were microwave ovens and receipt printers on cash registers. A large number of noise bursts were recorded in each measured

Snapshot It 211

* lo l2 l4 l6 rime (microseconde)

Fig. 2. Snapshot of a typical impulsive noise waveform measured in the 918 MHz band during a three-minute measurement run near an operating cash register. This waveform was recorded using a 1 ps/div horizontal sweep speed (40 ns bin duration) and the discone antenna.

The three horizontal sweep rates used in the measurements re- late to three different time resolutions per bin in the following manner:

1 ps/div (j 40 ns/bin 100 psldiv 4 ps/bin 10 ms/div e 400 ps/bin.

Measurement runs at each location in Sites A and B used all three oscilloscope sweep speeds, whereas only the 10 ms/div sweep speed was used in Sites C, D, and E.

111. MEASUREMENT RESULTS

A. Overview of Impulsive Noise Results

A snapshot of a typical measured waveform is shown in Fig. 2. This waveform was recorded at Site A near an operating cash register, and is a single swept waveform measured in the 918 MHz band with a l ps/div horizontal sweep speed (40 ns/bin). The number of waveforms measured during a three-minute measurement run depended on the number of impulsive events that occurred above the trigger level. For example, Fig. 2 is one of 523 snapshots (267,776 bins) recorded during a single three-minute measurement run at one location.

It is important to note the noise data measured with the oscilloscope’s peak detector (peak detectors and quasi-peak detectors have been used in the past to measure impulsive noise [6]-[ lo]) are worst-case amplitude measures. The digitizing oscilloscope quantized impulsive noise waveforms measured with a horizontal sweep speed of 1 ps/div into 40 ns time intervals (bins). If 50 ns constant amplitude bursts were present, the oscilloscope represented these pulses as 80 ns (two-bin) impulses. For the data measured with the 10 ms/div sweep speeds, it is possible that as many as 8,000 individual

frequency band during the operation of two cash registers. The measurements performed at Sites C and D indicate that

sources of impulsive noise in these environments were copy machines, printers, and elevator door switches. When the noise receiver was located in the vicinity of these sources, a large number of impulsive noise events were measured. Elsewhere in these buildings, very little impulsive noise was measured.

Noise generated by a microwave oven (located 15 meters from the receiver and behind a drywall partition) was detected at Site B in the 2.44 GHz band. Fig. 3 is one snapshot of the impulsive noise produced by the microwave oven. The maximum peak power received by the discone antenna was approximately -50 dBm. The spectrum of the noise generated by the microwave oven contained spectral lines separated by less than 200 Hz and had a bandwidth greater than 30 MHz. The noise bursts produced by the microwave oven had a period of 16 ms due to 60 Hz AC, and a duty cycle of approximately 50% (as indicated in Fig. 3). Most microwave ovens operate at a nominal frequency of 2.45 GHz, although this drifts over many tens of MHz in a few seconds [14]. Therefore, noise produced by the microwave oven can be modeled, to a first order, as a 2.45 GHz carrier modulated by a 60 Hz square-wave pulse train. More extensive modeling techniques are described in [14]. Noise from an operating microwave oven was also detected with the omnidirectional antenna in Site A with a received peak power level of -68 dBm (the microwave oven and receiver were separated by 50 meters and obstructed by a cinder-block wall and several rows of metal stock shelves).

B. Impulsive Noise Statistics

Since the measurement system recorded the maximum noise power level within each bin interval, information about the exact continuous distribution of the noise impulses is un- known. Only the peak amplitudes of the measured impulsive noise within a bin are known. However, this information is sufficient to find accurate peak amplitude probability distribution (PAPD) and cumulative distribution functions (CDF’s) of peak-noise pulse durations and burst spacings (interarrival time

BLACKARD et al.: MEASUREMENTS AND MODELS OF RADIO FREQUENCY IMPULSIVE NOISE

-a[ fllCPOWUE nr SITE B 2.44 8Hz 400 us Bin D u r a t i o n

Snapshot H40 hvp. Peak Pouer = -66.26 dBn

p 20 40 80 80 loo 120 140 180 180 zoo Time (mil l i inas)

Fig. 3. A snapshot of the noise produced by an operating microwave oven at Site B. The noise waveform was measured in the 2.44 GHz band. The microwave oven and receiver were 15 meters apart and separated by a drywall partition.

statistics) using the data measured with the fast sweep speed (40 ns/bin). The mean and standard deviation of all impulsive noise characteristics are also determined. Measurements using the slower sweep speed show the effects of noise sources with low repetition rates.

1) Amplitude Statistics: Amplitude statistics of impulsive noise are presented using the peak amplitude probability distribution (PAF’D) from measurements. Let Po denote a specific peak power level within a sampled bin for a given receiver bandwidth; then, we define the PAE’D(Po) to be the probability that a sampled peak power level exceeds Po,

PAPD(Po) = Probability(P 1 Po) (2) = 1 - PCDF(P,) (3 )

where PCDF(P,) represents the peak cumulative distribution function. Note our definition uses the peak values of power within an observation interval (bin). Our models guarantee a worst-case estimate of the noise properties, so conservative performance analysis may be carried out for indoor communication systems design.

The data measured with the fastest sweep speed provided the most accurate estimation of peak amplitudes of single pulses and was found to be accurate to within a couple of dB of the true average value within a bin. (This is expected since the bin duration represents an equivalent bandwidth which is approximately equal to the receiver bandwidth). For this reason, only the noise data measured with the 1 ps/div horizontal sweep speed in Sites A and B are used to compute the impulsive noise PAPD’s for each frequency band.

The technique for computing the PCDF( Po) in (2) and (3) is to determine the fraction of the total number of samples (bins) that have power levels less than Po, where Pmin 5 Po 5 P,,,. In the PAPD results presented in this paper, Pmin is set equal to the average thermal noise power of a noiseless (ideal)

DurnmyLood

0 10 20 30 40 50 60 70

Amplitude in dB above kTB

~

995

Sites A & B (All Locations)

Fig. 4. Typical peak amplitude probability distributions (PAPD’s) determined from all data recorded at Sites A and B using a 1 ps/div horizontal sweep speed. The average peak thermal noise level of the measurement system is shown by the “Dummy Load” curve.

measurement system for each measured frequency band [13]

where IC = 1.38 . 10-23J/K, To = 290 K, and B is the 3 dB bandwidth of the receiver system in Hz. The maximum value of peak power, P,,,, is set at a level 70 dB above the ideal average thermal noise power of the noise receiver in most cases. The maximum peak power levels in most cases did not exceed 70 dB above kToB (however, impulsive noise produced by the microwave oven, at a distance of 8.2 meters from the receiver at Site E, exceeded kToB by 77 dB).

Fig. 4 shows the PAF’D’s for the three bands measured in Sites A and B and the receiver thermal noise. The figure represents approximately 10,000 oscilloscope sweeps made in seven measurement locations, for an approximate total of 5,120,000 bins, each of 40 ns duration. The figure indicates that impulsive noise amplitude levels were significantly greater in the 918 MHz band than in the other measured bands. The tails (0.001% levels) of the PAPD’s shown in Fig. 4 indicate the maximum amplitude levels measured in the 918 MHz band were 19 and 20 dB higher than those measured in the 2.44 and 4 GHz bands, respectively.

One might (incorrectly) assume that impulsive noise energy is constant over a wide bandwidth. If this assumption is valid, then equal gain antennas will receive less energy at higher frequencies for a particular impulsive noise source. However, the 19 dB difference in the tails in Fig. 4 are more than can be explained by the 1/ f factor in free-space propagation loss over the range of 0.9-4 GHz. The free space path loss for frequencies of 2.44 and 4 GHz should only be 8.5 and 12.8 dB, respectively, higher than the free-space path loss at 918 MHz. The path loss differences predicted by an equal power wideband source differ greatly from the differences in the tails of the PAPD’s shown in Fig. 4. This suggests the power radiated by impulsive noise in retail stores and office buildings is not constant over wide bandwidths, and may be due to impulsive noise sources which are bandlimited or distributed (rather than point sources). Further research is needed to investigate this effect.

996

Average 1% h o b . Amp. (dB) a(dB) (de )

~ I ~ M H Z 18.87 6.86 40

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 11, NO. 7, SEPTEMBER 1993

O.M)I% hob. (dB)

63

2.44GHz 16.14 6.99 29 45

The averages (in dB) and standard deviations (in dB) of the peak amplitudes, along with the 1% and 0.001% probability levels of the noise peak amplitudes measured in each frequency band at Sites A and B, are listed in Table I. Although Fig. 4 shows that the difference in the tails of the impulsive noise PAPD’s between 918 MHz and 2.44 and 4 GHz were about 20 dB, the peak amplitude averages listed in Table I suggest amplitude levels measured in all three band were within 8 dB. The difference between the tails of the PAPD’s shown in Fig. 4 and the average peak amplitude levels in Table I are clearly due to the non-Gaussian peak amplitude distributions. This shows the average and standard deviation of impulse noise peak amplitudes are not sufficient to completely describe the amplitude distributions over the three bands measured.

2) Pulse Duration Statistics: The technique to determine pulse duration statistics requires two steps. First, a threshold level is specified so that only noise bursts above the threshold are considered in the statistical computations. The mean level of the peak bin amplitudes over an entire waveform, P p e a k ,

is used as a threshold level for pulse duration calculations for each waveform (the horizontal line on the graphs in Fig. 2 represents the mean level of the peak amplitudes, P p e a k ,

across a single snapshot). Then, the duration of pulses with amplitudes above the threshold level are determined for each snapshot, and statistics are assimilated over the ensemble of snapshots. The mean peak power level of each waveform was used as a threshold, since the average peak levels of all waveforms were found to have a very small variance at the fastest sweep speed.

Often, noise pulses with durations greater than a single bin width occurred for particular threshold settings. In such cases, the pulse was assumed to exist over an integer number of bins. A pulse was considered present when the peak power level of one or more consecutive bins exceeded the threshold level.

Typical pulse duration distributions (PDDs) for data measured in each frequency band are shown in Fig. 5 , and extensive data are presented in [17]. The PDD’s of data collected at all other sites are very similar to those shown in Fig. 5 , which suggests that pulse duration characteristics of impulsive noise are not dependent upon measurement location. The PDD’s shown in Fig. 5 indicate the pulse durations of impulsive noise bursts measured in all three frequency bands were comparable, although pulse durations in the 2.44 GHz band were slightly longer than in the 918 MHz and 4 GHz bands.

The PDD’s shown in Fig. 5 were compiled from the data measured with a 1 psfdiv sweep speed (40 ns bin duration).

IMPULSIVE NOISE PDD Typical Case (Site A)

T* - 40 ns

h

C

? 1 0 ,x .z 0.1 0” 2 y 0.01 U e

0.001

Pulse Duration (sec)

Fig. 5 . Typical pulse duration distributions (PDD’s) determined from the impulsive noise data recorded at Site A using a 1 ps/div horizontal sweep speed. These PDD’s were calculated using a threshold level equal to the average peak power of each measured waveform.

The bin widths of the slower two-sweep speeds are much wider than the minimum time resolution of the measurement system and have little significance in the interpretation of individual impulsive noise burst durations.

3) Pulse Spacing Statistics: Deriving pulse spacing statistics from the measured data requires care because the measurements with three different sweep speeds provide distinct bounds on the resolvable range of interarrival time statistics. For the statistical analysis presented in this paper, separate pulse spacing distributions were computed for measurements using each of the three sweep speeds.

The technique to determine the PSD for measured data using each sweep speed requires three steps. The first step is to set the pulse threshold level to the average waveform peak power level, P p e a k . Then, a pulse is defined as in Section 11-B. The third step is to determine the distribution of the time spacings between consecutive noise bursts which exceed the threshold. This is accomplished by calculating the times between two consecutive positive-going threshold crossings over each of the measured peak waveforms.

Fig. 6(a)-(c) show PSD’s of the data measured at Sites A-D in each frequency band with each of the three horizontal sweep speeds. These PSD’s were determined using a threshold level equal to the average peak power of the waveform, and are typical of the PSD’s calculated (with a threshold level equal to the average peak power) for each measurement site. As shown in [17, Appen. C], Fig. 6 is representative of a large number of locations in a particular building, although the spacings varied by almost an order of magnitude in some bands in some buildings at the 0.001% level.

Fig. 6(a) suggests distributions of spacing between consecutive impulses were similar, down to the 1% level, in each measured band when a 1 psfdiv sweep speed was used in the measurements. Fig. 6(b) indicates pulse spacings in the 918 MHz were closer than in the 2.44 GHz and 4 GHz bands when a 100 psfdiv sweep speed was used.

The PSD for the 2.44 GHz band in Fig. 6(c) is significantly different from the PSD’s for the other frequency bands. This is

-

BLACKARD et al.: MEASUREMENTS AND MODELS OF RADIO FREQUENCY IMPULSIVE NOISE

IMPULSIVE NOISE PSD All Measurements

T b 40 flS

2- z 0.1 n 2

e

0

0.01 U

0.001

Spacing Between Consecutive Pulses (sec)

(a)


Tb = 4.0 PS 100 z

0 9 10 h

m c

: : I

3 B 0.1 n P < 0.01

e

!4l

-

U

a 0.001


(b)


~b = 400 us 100

P 0 9 10 h

0,

$ 1 :: 2 B 0.1 x e 7 0.01

r

-

U

a 0.001

I '. \ "\

F - 91 I


(c)

Fig. 6. Pulse spacing distributions (PSD's) of impulsive noise measured at all sites with sweep speeds of (a) 1 hs/div, (b) 100 ps/div, and (c) 10 ms/div in each frequency band. A threshold level equal to the average peak power of each recorded snapshot was used to determine these distributions.

a result of a large number of impulsive noise events recorded in the 2.44 GHz band during the use of microwave ovens in the measurement sites. Impulsive noise produced by microwave ovens is distinct. It appears at baseband as a pulse train with a period of 16 ms. The PSD for the 2.44 GHz band in Fig. 6(c)

991

indicates the presence of microwave ovens in the measured buildings. The vertically sloped portion of the 2.44 GHz PSD is located at approximately 16 ms.

C. Impulsive Noise Models

A simple mathematical modeling technique is used to com- press the graphical representations of the impulsive noise distributions described in the previous four sections and to facilitate their use in future system designs.

The modeling technique uses a piecewise-linear approximation to the true impulsive noise distribution, which is determined by using the data processing techniques in Section 111-B. True empirical distributions are sampled at the loo%, 50%, lo%, 1%, 0.1%, 0.01%, and 0.001% levels, and the corresponding abscissa values are tabulated. For example, if a PAPD is to be modeled, then seven samples (100-0.001%) are made and their corresponding amplitudes above thermal noise are tabulated. By using a logarithmic probability scale and passing straight-line segments through the tabulated points, a simple and accurate approximation to the true distribution can be achieved.

Third-order and fifth-order least squares modeling techniques were also examined and compared to the piecewise- linear modeling method. The least squares approximations did not illustrate accuracies significantly greater than the piecewise-linear approximation and suffered severely from roundoff error (16-place coefficient accuracy was necessary to reasonably approximate the impulsive noise distributions).

Piecewise-linear parameters for modeling the first-order PAPD's, PDD's, and PSD's are listed in Tables II(a)-(c). Joint statistics are not considered in this work. More complicated mathematical models for narrowband impulsive noise amplitude distributions (APD's) were developed by Middleton [ 151 and used to analytically determine the effects of impulsive noise amplitude on various signal detection techniques in [16]. However, [16] does not consider the effects of pulse duration or pulse spacing intervals.

In order to completely understand how impulsive noise affects communication system performance, the amplitude, pulse duration, and pulse interval statistics must be employed. Computer simulation is a viable method for recreating these noise statistics and for determining their effects on indoor wireless communications. The simple models of impulse amplitude, duration, and interval statistics presented in this paper have been used to create sequences of computer-simulated impulsive noise which have similar statistical distributions as measured data [17]. Fig. 7 compares the PDD's for measured and simulated impulsive noise in the 4 GHz band at Site A. This figure illustrates how the simple PDD models presented in this paper can be used to accurately recreate impulse duration statistics. The models for amplitude and interval statistics can also be used to accurately simulate impulsive noise with statistics similar to those measured [17].

D. Results of Specific Noise Source Measurements

This section presents results obtained from the processed impulsive noise data recorded at Site E. Three noise sources

998 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 11, NO. 7, SEPTEMBER 1993

0.001 %

TABLE 11

(a) PAPD’s, (b) PDD’s, (c) PSD’s PIECEWISE-LINEAR MODELING PARAMETERS FOR MEAS~JRED

63 45 43 1

11% 1 4 0 I 2 9 I 2 3 I

I00 9%

50 %

10 %

1 90

0.1 %

0.01 %

10.1% I 48 I 31 1 34 I

80 80 80

120 120 120

240 280 240

440 600 400

720 1440 640

1400 2960 1040

918 MHz 2.44 GHz (4 (ns)

100 % 80 80

50 % 200 200

4.0 GHz (ns)

80

200

1 %

0.1 %

0.01 %

0.001 %

600 840 600

960 1880 840

1920 4200 1160

7240 12,800 1600

were examined: a pay-per-copy photocopier (Savin 7050), an elevator (motor and door-opening switches), and a microwave oven (Kenmore Model #565). Results are presented in the form of PAPD’s, PDD’s, and PSD’s as functions of noise source.

I ) Amplitude Probability Distributions of Specific Sources: Fig. 8(a)-(c) show the PAPD’s of impulsive noise produced by the photocopier (6.1 meters from the receiver), elevator (the receiver, located in the corridor adjacent to the elevator,

3 a 0.1 n

t 0

E O.O1 : 0.001

1

IMPULSIVE NOISE PDD Site A - 4.0 GHz

1 00 z 9 l o 0 m

h

Pulse Duration (sec)

Fig. 7. PDD’s of measured and simulated impulsive noise (for grocery store environment) for 4 GHz. A threshold level equal to the average peak power of each snapshot was used to determine these distributions.

was 2.4 meters from the elevator door), and microwave oven (8.2 meters from the receiver). By comparing Fig.8 (a) and 8(b) and noting T-R separations for measurements made during the operation of the copier and elevator, it is evident the impulsive noise produced by the photocopier was much more significant than impulsive noise generated by the elevator door switches. Notice that the tail of the photocopier impulsive noise PAPD at 4 GHz is only 7 dB lower than the PAPD tail at 918 MHz, where as (5 ) would predict a 12.8 dB weaker signal for the 4 GHz tail. This suggests the impulsive noise power produced by photocopiers is not constant over a wide band. It is possible that the metal case of the photocopier provides better RF shielding at 918 MHz and 2.44 GHz than at 4 GHz. There are several small openings and cracks in the metal case (such as the slots for the paper trays), which may allow the high-frequency (>918 MHz) content of the impulsive noise produced inside the photocopier to radiate more freely than the lower frequencies. In this case, the effective radiated power spectrum from the photocopier will not be constant over all frequencies.

No significant impulsive noise was detected in the 918 MHz and 4 GHz bands during the operation of the microwave oven. For this reason, Fig. 8(c) only shows the PAPD of noise produced by the microwave oven in the 2.44 GHz band. Also, note the very high amplitude levels measured. Peak power levels of -27 dBm were measured at the receiver input using a 12 dBi gain helical antenna.

The average peak amplitudes, standard deviations, and 1 and 0.001% probability amplitude levels as functions of frequency and noise source are listed in Table 111.

Additional measurements were made to determine if the photocopier in Site E could be modeled as a point source. Seven one-minute measurement runs were made at 5.5, 7, 8.5, 10, 11.5, 13, and 16 meters from the photocopier, with a line- of-sight path inside a hallway. The measurements were made in the 918 MHz band using a directional helical antenna. A horizontal sweep speed of 10 ms/div was used so that several impulses (as many as 400) could be captured per waveform.

One way to model the photocopier as a point (noise) source is to determine the received impulsive noise power from the

BLACKARD et al.: MEASUREMENTS AND MODELS OF RADIO FREQUENCY IMPULSIVE NOISE 999

Source - Photocopier 1-R Separotion - 6 1 m

0 100 v1

U

9 10

W U

c = 1

27 :j 0.1

2 5 0.01

0,

c W

a" 0.001 0 10 20 30 40 50 60 70

Amplitude in d 8 above kTB

(a)

0 100 v1

U

2 10 h

W U 3 + ' a ' E

Source - Elevator Switch T-R SeporotlOn - 2 4 m

91 8 MHz 2 4 4 GHz

I ' 5 0.1

P i i 001 W ? I 2 0 001 \ , . I . I 1 I

0 10 20 30 40 50 60 70

Amplitude in dB above kTB

( b)

Source - Microwave Oven (2.44 GHz) T-R Ssparotion - 8.2 rn

c

% 1

3

n

4

5 0.1

5 0.01

0

2

W

k a 0 . 0 0 1 f ' " " " " ' ' ' ' '

0 10 20 30 40 50 60 70 80

Amplitude in dB obove kTB

(c)

Fig. 8. Peak amplitude probability distributions (PAPD's) measured in each frequency band during operation of the (a) photocopier, (b) elevator, and (c) microwave oven at Site E.

photocopier as a function of T-R separation. Generally, the received power will exhibit a log-distance relationship specified by a path loss exponent [1]. For free-space transmission, the path loss exponent n is equal to 2.

PAPD's were computed for each of the seven T-R separations using the technique described in Section 111-A. The PAPD curves all had the same general shape, which indicates the shape of the amplitude distribution does not depend upon

TABLE I11 AVERAGE PEAK AMPLITUDES, STANDARD DEVIATIONS, AND 1% AND 0.001%

PROBABILITY PEAK AMPLITUDE LEVELS AS FUNCTIONS OF FREQUENCY AND NOISE SOURCE

I Photocopier I Elevator I 1 TI- I I I I

918 M H z I Mean (dB) 1 15.9 I 14.9 I No Noise

Received Powervs. T-R Separation Noise Source - Photocopier

, . , . . . . . . , . , . .

. . . . . .

2 10

-15.03 1 ; b ; b ; ' T-R Separation (m)

Fig. 9. Scatter plot of normalized received noise power from an operating photocopier as a function of T-R separation. The slope of regression line drawn through the data points represents the best-fit path-loss exponent, R, for the measured data.

T-R separation. To explore path loss behavior of the most significant impulse noise events, each PAPD was sampled at the 0.1% probability level and the corresponding amplitudes were normalized so that the maximum peak-amplitude level at the 0.1% probability level was 0 dB. Fig. 9 shows a scatter plot of a normalized received amplitude level as a function of T-R separation. A regression line was passed through the measured sample points to determine the path loss exponent, which is the slope of the line. The value of n determined from the measurements is 1.7. This value is slightly lower than the free-space path loss exponent (n = 2). Similar path loss exponents were measured in [4] along hallways; hence, it seems reasonable to model the photocopier as a point source of impulsive noise having a d-1.7 falloff of power with distance d.

2) Pulse Duration Distributions of Specific Noise Sources: PDD's of the impulsive noise produced by the photocopier and elevator switches indicate pulse durations are similar in all bands. However, the pulse duration distributions of

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 11, NO. 7, SEPTEMBER 1993

U (ns) 1% Prob (ns)

0.001% Prob (ns)

TABLE 1V AVERAGE PULSE DURATIONS, STANDARD DEVIATIONS, AND 1% AND 0.001%

PROBAslLlTY IMPULSE DURATIONS AS FUNCTIONS OF NOISE SOURCE AND FREQUENCY. A THRESHOLD LEVEL EQUAL TO THE AVERAGE PEAK

DETERMINE THESE STATISTICS POWER OF EACH RECORDED SNAPSHOT WAS USED TO

259 102 Detected

I ,480 560

3,960 920

the photocopier impulsive noise suggest somewhat longer durations than those for the elevator door opening switches.

The PDD of the measured microwave oven noise indicates pulse durations of 8 ms. This was also observed during the measurements. The noise waveforms produced by the microwave oven had a period of 16 ms, due to the 60 Hz switching power supply used to drive the magnetron, and a duty cycle of approximately 50%. This is also illustrated in Table IV, which lists average pulse durations and standard deviations of the pulse durations as functions of noise source and frequency.

3) Pulse Spacing Distributions of Specijic Noise Sources: Table V lists the average spacings between consecutive noise bursts and standard deviations of the spacing interval as functions of noise source and frequency. This table lists pulse spacing statistics found using a threshold level equal to the average peak power of each measured waveform. This table indicates average pulse spacings of impulsive noise produced by the photocopier and elevator switches are similar in all bands. Table V and the PSD of the microwave noise in Fig. 6(c) indicate a 16 ms spacing between consecutive noise bursts generated by the microwave oven, which is caused by 60 Hz switching.

4) Impulsive Noise Models: The modeling technique described in Section 111-C has also been used to mathematically model the noise characteristics of the photocopier, elevator, and microwave oven measured in Site E, and are listed in [17, Appen.].

IV. CONCLUSIONS This paper has statistical results of impulsive noise measure-

ments within five retail stores and office buildings. The work has shown that devices with electromechanical switches (elec- tric motors in elevators, refrigeration units, copy machines, printers, etc.) are principal sources of impulsive noise in retail and office environments in the low microwave regime.

Measured noise peak amplitudes were consistently higher in the 918 MHz band than in the other two measured bands, excluding measurements made in the 2.44 GHz band during the operation of microwave ovens. We speculate that this is due to higher path losses at higher frequencies and also due to spurious interference from mobile radio and paging broadcasts.

The pulse duration statistics compiled from data at all measurement sites indicate distributions in all three frequency bands are comparable. Average pulse durations in all three frequency bands range from 120 to 150 ns (for a threshold level equal to the average peak power level) with standard deviations on the same order.

The pulse spacing statistics calculated from the data at all locations indicate that average spacings between consecutive noise pulses are similar in each frequency band. However, average pulse spacings are a function of bin duration. Calcu- lated average pulse spacings can range from a few hundred nanoseconds to several milliseconds, depending upon the bin duration. Thus, a comprehensive statistical model of pulse spacing that is independent of temporal resolution must be developed to completely model impulsive noise.

ACKNO w LEDGMENT

The authors thank S. Seidel and M. Keitz for their assistance in the development of the measurement system and data collection during the measurement campaign. The authors gratefully acknowledge NCR Corporation for funding this research and permitting them to publish this work. The authors also thank B. Tuch of NCR Corporation for his helpful editorial comments and suggestions.

REFERENCES

[ l ] T. S. Rappaport, S. Y. Seidel, and K. Takamizawa, “Statistical channel impulse response models for factory and open plan building radio communication system design,” IEEE Trans. Commun., vol. 39, no. 5, pp. 794-807, May 1991.

BLACKARD ef al.: MEASUREMENTS AND MODELS OF RADIO FREQUENCY IMPULSIVE NOISE 1 0 0 1

[2] T. S. Rappaport, “Characterization of UHF multipath radio channels in factory buildings,” IEEE Trans. Anten. and Propagat., vol. 37, no. 8,

[3] D. C. Cox, “Universal portable radio communications,” IEEE Trans. Vehic. Technol., vol. VT-34, no. 3, pp. 117-126, Aug. 1985.

[4] T. S. Rappaport and D. A. Hawbaker, “Wide band microwave propagation parameters using circular and linear polarized antennas for indoor wireless channels,” IEEE Trans. Commun., vol. 40, no. 2, Feb. 1992.

[5] J. D. Parsons and J. G. Gardiner, Mobile Communication Systems. Lon- don: Blackie and Son, 1989.

[6] J. D. Parsons and A. Sheikh, “The characterization of impulsive noise and considerations for a noise-measuring receiver,” Radio and Electron. Engineer, vol. 49, no. 9, p. 468.

[7] R. A. Shepherd, “Measurements of amplitude probability distributions and power of automobile ignition noise at HF,” IEEE Trans. Vehic. Technol., vol. VT-23, no. 3, pp. 72-82, Aug. 1974.

[8] G. L. Maxim, H. P. Hsu, and P. W. Wood, “Radiated ignition noise due to the individual cylinders of an automobile engine,” IEEE Trans. Vehic. Technol., vol. VT-25, no. 2, pp. 33-38, May 1976.

[9] C. Egidi and E. Nano, “Measurement and suppression of VHF radio interference caused by motorcycles and motor cars,” IRE Trans. Radio

pp. 1058-1069, Aug. 1989.

Freq. Interfer., pp. 2-10, 1961. 1101 E. Skomal, Man-Made Radio Noise. New York: Van Nostrand Rein- .~

hold, 1978. [ll] T. S. Rappaport, “Wide-band Test Antennas,” RF Design, pp. 37-41,

Apr. 1988. (121 J. D. Kraus, Antennas. [13] E. Skomal and A. Smith, Jr., Measuring the Radio Frequency Environ-

ment. [14] J. Y. C. Chia, “Interference characteristics of microwave ovens in indoor

radio communications,” IEEE 802.1 1 Tech. Commit. Pub. 11/91 -52, May 1991.

[15] D. Middleton, “Statistical-physical models of man-made radio noise- Part I: First-order probability models of instantaneous amplitude,” Off. of Telecommun. Reo. OT 74-36. Aor. 1974.

New York: McGraw-Hill, 1988.

New York: Van Nostrand Reinhold, 1985.

A. D. Spaulding and D. Middleton, “Optimum reception in an impulsive interference environment-Part I: Coherent detection,” IEEE Trans. Commun., vol. COM-25, no. 9, pp. 910-923, Sept. 1977. K. L. Blackard, “Measurements and models of radio frequency impulsive noise inside buildings,” Masters Thesis, Virginia Polytech. Instit. and State Univ., Oct. 1991. K. L. Blackard, T. S. Rappaport, and C. W. Bostian, “Radio frequency noise measurements and models for indoor wireless communications at 918 MHz, 2.44 GHz, and 4.0 GHz,” in Proc. IEEE Int. Con$ Commun., Denver, CO, June 1991, pp. 28-32.

Theodore S. Rappaport (S’83-M’84-S’85- M’87-SM’gl) was born in Brooklyn, NY, on November 26, 1960. He received the B.S.E.E., M.S.E.E. and Ph.D. degrees from Purdue University in 1982, 1984, and 1987, respectively.

In 1988, he joined the Electrical Engineering Faculty of Virginia Tech, Blacksburg, where he is an Associate Professor and Director of the Mobile and Portable Radio Research Group. He conducts research in mobile radio communication system design and RF propagation prediction

through measurements and modeling. He guides a number of graduate and undergraduate students in mobile communications and has authored or coauthored more than 70 technical papers in the areas of mobile radio communications and propagation, vehicular navigation, ionospheric propagation, and wideband communications. He holds a U.S. patent for a wideband antenna and is co-inventor of SIRCIM, an indoor radio channel stimulator that has been adopted by over 80 companies and universities. In 1990, he received the Marconi Young Scientist Award for his contributions in indoor radio communications, and was named a National Science Foundation Presidential Faculty Fellow in 1992. He serves as Senior Editor of the IEEE JOURNAL ON SELEcrEo AREAS IN COMMUNICATIONS. He is a Registered Professional Engineer in the State of Virginia and is a Fellow of the Radio Club of America. He is also President of TSR Technologies, a cellular radio and paging test equipment manufacturer.

Charles W. Bostian (S167-M’67-SM’77-F’92) was born in Chambersburg, PA, on December 30, 1940. He received the B.S., M.S., and Ph.D. degrees in electrical engineering from North Carolina State University in 1963, 1964, and 1967.

After a short period as a Research Engineer with Corning Glassworks and two-years service as a US. Army Officer, he joined the Virginia Tech faculty in 1969 and is currently Clayton Ayre Professor of Electrical Engineering. From 1972 through 1988, he headed Vireinia Tech’s Satellite Communication

Group and, since 1991, he has been Executive Director of the University’s Center for Commercial Space Communications, On leave during the 1989 calendar year, he served as an IEEE Congressional Fellow on the staff of Representative Ritter, working on legislative issues related to the US. electronics industry and economic competitiveness. His primary research interests are in radio wave propagation and satellite communications. He has been principal investigator of sponsored research projects totaling approximately $3.8M and has published approximately 35 journal articles. He teaches primarily in the areas of communications and electromagnetics. His teaching has been recognized by a number of awards, and he is a member of the Virginia Tech Academy of Teaching Excellence. He is the co-author of two textbooks, Sold State Radio Engineering and Satellite Communications, both published by Wiley. He is Chair of the IEEE-USA Engineering R&D Policy Committee, and serves on the IEEE-USA Technology Policy Council and the Congressional Fellows Committee. He is Associate Editor for Propagation of IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION.

Kenneth L. Blackard (S’90-M’91) was born in Mount Airy, NC, on July 5, 1967. He received the B.S.E.E. and M.S.E.E. degrees from Virginia Polytechnic Institute and State University in 1989 and 1991, respectively.

From 1990 to 1992, he worked with Profes- sor T. S. Rappaport in the Mobile and Portable Radio Research Group where he did research on radio-frequency impulsive noise and radio wave propagation. In November 1992, he joined the Fed- era1 Bureau of Investigation where he is currently

employed as a Research Engineer. His research interests are RF propagation, wireless communications, and antennas. He is a member of the Eta Kappa Nu, Phi Eta Sigma, and Phi Kappa Phi honor societies.

measurements and models of radio frequency...

Documents