Characterizing the 2.4 GHz Spectrum in a Hospital Environment: Modeling and Applicability to Coexistence Testing of Medical Devices

Abstract

The increasing use of shared, unlicensed spectrum bands by medical devices and nonmedical products highlights the need to address wireless coexistence to ensure medical device safety and effectiveness. This paper provides the first step to approximate the probability of a device coexisting in its intended environment by providing a generalized framework for modeling the environment. The application of this framework is shown through an 84-day spectrum survey of the 2.4–2.48 GHz industrial, scientific, and medical band in a hospital environment in the United States. A custom platform was used to monitor power flux spectral density and record received power. Channel utilization of three nonoverlapping channels of 20 MHz bandwidth—relative to IEEE 802.11 channels 1, 6, and 11—were calculated and fitted to a generalized extreme value distribution. Low channel utilization was observed (<10%) in the surveyed environment with sporadic occurrences of higher channel utilization (>50%). Reported findings can be complementary to wireless coexistence testing. This paper can provide input to the development of a consensus standard for wireless device coexistence test methods and a consensus document focused on wireless medical device coexistence risk management.

I. Introduction

MEDICAL device manufacturers are integrating WLAN and Bluetooth technology into medical devices to spur innovation in healthcare. These wireless technologies use the 2.4 GHz industrial, scientific, and medical (ISM) unlicensed band, which shares spectrum with other wireless devices, increasing the likelihood of communication loss and errors. Unlike spectrum bands that are dedicated for medical use, such as the wireless medical telemetry service¹ and the medical implant communications service,² the ISM band was intended for a broad range of applications (e.g., industrial and domestic) that generate and use locally radio frequency (RF) energy [1]. The fact that the ISM band is unlicensed, coupled with the availability of technologies it accommodates such as WLAN and Bluetooth, made it a popular choice for an increasing number of wireless-enabled medical devices. For example, it is reported in [2] that a majority of wireless medical devices cleared for marketing by the U.S. Food and Drug Administration (FDA) use either Bluetooth or WLAN. A challenge of incorporating wireless communication into a medical device is ensuring reasonable safety and effectiveness [3]. As a result, the FDA issued a guidance document regarding the use of wireless RF technology [4]. The recommendations include that medical device manufacturers should address wireless coexistence in their pre-market submission (application for permission to market the device). The guidance defines wireless coexistence as the ability of one wireless system to perform a task in a given shared environment where other systems (in that environment) have an ability to perform their tasks and might or might not be using the same set of rules. However, there are currently no standardized test methods for such an assessment, with the result that many wireless coexistence tests are performed ad hoc.

A. Coexistence: Standardization and Methods

At the time of publication of this paper, work to develop a standardized process for an RF wireless coexistence assessment was underway by Subcommittee 7 (Spectrum Etiquette) of American National Standards Institute (ANSI)—accredited standards committee C63. The future standard, currently in draft, is designated C63.27, standard for evaluation of wireless coexistence [5]. ANSI is a private, nonprofit organization that oversees the development of voluntary consensus standards in the United States. ANSI also coordinates U.S. standards with international standards. After a wireless device is tested for coexistence, an optional next step is to derive the approximate probability of the device coexisting in its intended environment. This paper provides the first step to approximate this probability by providing a generalized framework to model the intended environment of the device under test. Experimental data are provided to show the application of this framework.

Several methods have been suggested in the literature and are under review for inclusion in C63.27 [6]. One uses RF components (e.g., combiners, couplers, and attenuators) to establish a wired communication link between medical device wireless nodes so that interference can be introduced through the use of signal generators or by way of an actual wireless network. This form of testing is referred to as conducted testing and has the downside of requiring access to the medical device’s antenna ports. Radiated testing does not require access to device antenna ports and can be performed in an anechoic chamber [2], two smaller anechoic chambers [7], [8], or in alternative low-noise environments [9]. Similar radiated methods of characterization are known as over-the-air testing in several domains, such as cellular networks research [10]. Additionally, the use of reverberation chambers to introduce electromagnetic fields that emulate disturbances in realistic environments has been suggested [11].

Although suggested coexistence testing methods differ, they all are used to attempt to characterize the medical devices in terms of the basic physical layer coexistence parameters of distance, frequency, and time and how the higher open systems interconnection (OSI) layers mitigate interference. Notably, the physical layer of a receiving node could undergo blocking—which cannot be mitigated in higher OSI layers—due to the proximity of nearby transmitters on the same or adjacent RF channels. When experiencing interference, a medical device signal-to-interference-plus-noise ratio (SINR) decreases. This can be emulated by controlling distance, either physically or by varying transmission power/attenuation, to provide insight about expected performance of the medical device relative to its separation distance from an interfering network. The interferer’s operating frequency can be set on a channel that overlaps the one used by the medical device (i.e., cochannel) or on an adjacent channel that could cause interference as a result of imperfect filtering. The result would be a decrease in SINR. Both scenarios can be accounted for during testing by controlling the interfering node’s RF parameters. The probability of successful packet transmission increases when the channel occupancy time (i.e., channel utilization (CU) or duty cycle) decreases. The CU threshold at which a medical device can successfully achieve its wireless functions can be determined by varying the interferer’s throughput.

B. Contribution

The main contribution of this paper is to provide a starting point for finding a robust statistical model of the electromagnetic environment in which a medical device would typically operate. Such statistics are reported in a way that could be meaningful to the healthcare community. Coexistence testing protocols address devices under test that use a broad range of standardized or proprietary wireless protocols. Regarding causes of interference, the WLAN family of protocols, operating on 20 MHz bandwidths, is capable of generating high CU of 80–90% [12]. Therefore, they are the most significant potential cause of interference for devices sharing the spectrum in their vicinity [13]. Consequently, CU of WLAN³ channels 1, 6, and 11 was measured and fitted to a generalized extreme value (GEV) distribution. CU quantifies the wireless spectrum occupancy, which is associated with the likelihood that new devices, attempting to use the spectrum, have for wireless coexistence. Results are presented to highlight the temporal distribution of CU relative to the time of day. An example of expected coexistence behavior of a wireless device in the surveyed environment is discussed in Section IV-C with regard to the relationship between the allowed time window of the device’s wireless functions and measured CU.

The remainder of this paper is organized as follows. Section II presents the background on previous spectrum surveys and the methods used. Section III describes the location, experimental setup, and methodology of our spectrum survey measurements. Section IV presents experimental results that can be used by the healthcare community. Section V concludes the paper.

II. Related Work

Spectrum surveys have attracted the interest of a wide variety of groups, including government agencies [14], [15], corporations [16], and academics [17]. The National Telecommunications and Information Administration (NTIA) of the United States Department of Commerce conducted several outdoor broadband spectrum surveys in the late 1990s. Surveyed locations included Denver, CO; San Diego, CA; Los Angeles, CA; and San Francisco, CA. Some surveys have been repeated (e.g., San Diego and Denver), and additional ones have been conducted in Chicago [15]. It was concluded that activities in the 2.4 GHz ISM band were unique to the measurement locations and attributed to background RF radiation generated by ISM devices and microwave ovens.

Microsoft Spectrum Observatory [16] was launched in response to the NTIA Spectrum Monitoring Pilot Program Notice of Inquiry [18]. Bands between 30 MHz and 6 GHz are monitored at locations distributed worldwide using Universal Software Radio Peripheral devices. Collected data are centrally stored and processed for visualization through the Windows Azure Cloud and made freely available to the public. Main output parameters include observed power density, utilization, and spectrograms, all of which can be obtained dynamically online. In July 2007, the Illinois Institute of Technology (IIT) started a permanent spectrum monitoring system in Chicago, IL. The system can interface with various data collection hardware. Some use a scan-based approach, offering wider observation bandwidth at the expense of lower time resolution. Others use a higher sampling rate focused on narrow bands (i.e., time-domain measurement of electromagnetic disturbances [19]), thus generating high time resolution measurements. The project has expanded to incorporate multinational collaborators, including IIT, Chicago, IL, USA; Virginia Tech (VT), Blacksburg, VA, USA; Turku University of Applied Sciences, Turku, Finland; University of Oulu, Oulu, Finland; and VTT Technical Research Centre, Finland [17]. Storage and data analysis aspects of the project are reported in [20].

Several researchers performed spectrum surveys in hospitals in the 1970s, including McDonnell Douglas Astronautics Co.-East under contract to the FDA Bureau of Medical Devices.⁴ McDonnell Douglas used the survey data to draft a medical device EMC standard [14], which includes citations of spectrum measurements made by other researchers. In 2003, researchers used a spectrum analyzer to survey several locations at two hospitals in Virginia, including the intensive care unit (ICU) and radiography units, over several days [21]. High disturbance levels were reported to be rare. Characteristics of wireless channels (i.e., path loss and power delay profile) in several bands used in medical applications were also surveyed in a hospital room in Japan [22]. In 2009, hospital surveys were reported in Switzerland [23] and Oklahoma, USA [24]. The former found evidence of rare electric field levels exceeding the 3 V/m immunity test level specified by IEC standard 60601-1-2 [25] for medical equipment that is not life supporting, while the latter did not. In addition to WLAN and Bluetooth signals, microwave oven radiation was observed at Kyoto University Hospital in 2013, as detailed in [26]. More recently, multiple surveys of hospitals in Finland [27]–[29] and Italy [30] were reported in the literature. Researchers adopted similar observation methods by deploying a spectrum analyzer at various locations in investigated hospitals. Received RF power was recorded and used to report CU. Findings were analogous and included low spectrum occupancy observations. A short-term survey of an ICU—conducted using a vector signal analyzer (VSA)—was used to evaluate Bluetooth Low Energy (BLE) for wireless coexistence in [31].

To the best of the authors’ knowledge, spectrum surveys of hospital environments lasting as long as the study detailed in this paper are yet to be reported in the literature. The integration of data collection equipment and a supercomputer processing platform allowed for efficient and flexible analysis of the large volume of data collected over nearly three months. A 28-day subset of survey data in this paper has been analyzed and published in [32].

III. Setup and Methodology

The surveyed location for the work detailed in this paper is the Children’s Hospital, equipped with 356 beds, at the University of Oklahoma (OU) Medical Center campus in downtown Oklahoma City, OK, USA. The combined facilities comprise the largest hospital in the state of Oklahoma. A vector signal analyzer (VSA) was used as the measurement hardware. The VSA has an average noise level of 157 dBm/Hz, 80 dB spurious-free dynamic range, and 50 MHz instantaneous bandwidth (at 3 dB). Test equipment was installed in a networking equipment cabinet in a 24/7 operational postsurgery recovery room (RR), where it was connected to a nearby 9 dBi omnidirectional antenna using a low-loss cable fed through a dropped ceiling. Signal loss caused by cable length and antenna gain was accounted for in data collection software. The RR is equipped with 16 beds separated by curtains. The antenna was placed at the RR entrance. Subject to hospital staff assistance, the survey lasted for 84 days, commencing December 4, 2014 and ending February 25, 2015.

Data collection software developed at OU [33] was used to collect dBm power measurements by scanning the wireless spectrum between 2.4 and 2.48 GHz. Instantaneous power measurements without averaging were acquired on a total of 1993 frequency bins, each having 40 kHz bandwidth, during a total dwell time of 4 ms. Random processing and disc I/O time increased average spectrum sweep capture time to approximately 12 ms. Measured dBm power samples were logged in text files, wherein each line represented one spectrum sweep. To facilitate data storage, a daily directory was created and power samples in 1-min intervals were grouped into a single subdirectory. Approximately 6.5 TB of data were collected and stored at the Tandy Supercomputing Center in Tulsa, OK, where a parallel program was developed for fast data processing.

A threshold is required for making a decision as to whether a channel is active or inactive at any given time. Methods for selecting an appropriate decision threshold T have been investigated extensively and reported in the literature. One such method suggests simply selecting T by visual inspection of power measurements and then deciding on a level that separates activity patterns from noise [29]. Further automation of the threshold selection process can be done by inspecting noise measurements to find the mean and maximum recorded noise values. Afterward, T can be considered to equal the maximum noise value, thus minimizing the false detection that would occur if noise samples were identified as genuine activity. Alternatively, T can be calculated as several dB above the mean noise value. For example, the International Telecommunication Union (ITU) spectrum occupancy measurements and subsequent evaluation report suggest that the threshold should be at least 3–5 dB above the noise level [34]. In this method, the noise variance contributes to false detection of noise samples as activity. In this study, a compromise between the “maximum noise value” and the “mean plus several dB” methods was adopted by selecting a given probability of false detections and calculating the value of T accordingly. Several measurement windows of 1-min duration were examined empirically. A high-activity window was then selected to derive a probability distribution function (PDF) of power values. Because noise samples are considered normally distributed, the threshold was fixed at two standard deviations from the noise mean, which is equivalent to 95% confidence in accurate activity detection. We assumed that the noise threshold was constant throughout the survey period. Consequently, the decision threshold was fixed at T = −79.84 dBm.

CU was used as a measure of spectrum occupancy in WLAN channels 1, 6, and 11 with 20 MHz bandwidth centered on 2412, 2437, and 2462 MHz, respectively. It was demonstrated previously in [32] that these three channels are active in the investigated environment. Other wireless systems that operate in the 2.4 GHz ISM band include Bluetooth and ZigBee. Bluetooth uses adaptive frequency hopping to avoid interference on any of its 79 channels of 1 MHz bandwidth. BLE uses 37 channels of 2 MHz bandwidth. ZigBee uses 16 nonoverlapping channels of 2 MHz bandwidth and typically uses much lower transmission power than WLAN. Consequently, none of these systems would have contributed significantly to the reported CU values. However, acquiring spectrum measurements at a frequency resolution higher than the bandwidth of either WLAN, Bluetooth or ZigBee (40 kHz in comparison with 20, 1, and 2 MHz, respectively) allows capture of such activity whenever it occurs in the vicinity of the monitoring antenna within a given spectrum sweep.

CU is the fraction of time a given channel is detected as active by the way of observing power values exceeding the decision threshold. This method conforms with the ITU definition of frequency band occupancy [34]. CU values were calculated using 1-s time resolution mapping variations in spectrum occupancy throughout the day. To do so, we divided each minute of collected data into 60 equal parts, assuming that power measurements were taken at a constant rate over a 1-min period. After applying threshold T to power measurements of a given WLAN channel C collected at time instance n, we constructed a binary matrix ${𝒳}_C^{s\times b}(n)$, where s (i.e., the number of rows) is the count of spectrum sweeps collected during 1 s; b = 500 (i.e., the number of columns) is the number of 40 kHz subchannels in 20 MHz; and x_i,j is the binary result of comparing the corresponding power measurement with T, yielding “0” given the measured power is < T, otherwise yielding “1”. Consequently, CU for time instance n is

$${\Phi }_C(n)=\frac{1}{sb}\sum _{i=1}^s\sum _{j=1}^b{x}_{i,j}.$$ (1)

CU quantifies the percentage of time the channel is occupied, which is a fundamental parameter for investigating and testing for coexistence [2], [9], [35]. It should be noted that our measurements, and others reported in literature, were performed at one location per survey. Wireless traffic sources contributing to calculated CU values are those near the data collection antenna. The unlicensed character of the 2.4 GHz ISM band, spectrum management practices of hospital staff, and human and machine demand for offered wireless services are all factors contributing to the uniqueness of each spectrum survey.

IV. Results

The dataset obtained when the spectrum survey concluded includes approximately 6.3 M Φ_C samples for each of 802.11 channels 1, 6, and 11. Φ_C is a discrete value that quantifies CU during time instance n as defined in (1). A sample of Φ_{1,6,11} variations in a 24-h period is depicted in Fig. 1. It can be seen that Φ remains near a minimal value for the majority of the day, with only sporadic occurrences of high activity. In this section, we present our findings relative to the probability distribution that best fits the observed Φ values. Furthermore, even though Φ typically remains at the low end of possible values, we highlight the cases of observed high spectrum activity and elaborate on instances when high values occurred during the study.

Fig. 1.

CU (Φ) variations on WLAN channels 1, 6, and 11 during 24 h. We notice that occurrences of high values are sporadic and are not concentrated in a specific time window.

A. CU Distribution

In addition to model fitting, empirical PDFs of Φ are illustrated in Fig. 2 for WLAN channels 1, 6, and 11. The GEV distribution was found to accurately fit the data based on Bayesian information criterion when compared with a group of alternative distributions previously reported in the literature as candidates for modeling CU (e.g., beta [36] and t-location scale [30]). The GEV distribution is given by the density function

$$f\left(\Phi \mid {\lambda }_1,{\lambda }_2,{\lambda }_3\right)=\left(\frac{1}{{\lambda }_3}\right)\text{exp}\left(-{\left(1+{\lambda }_1\frac{\Phi -{\lambda }_2}{{\lambda }_3}\right)}^{-\frac{1}{{\lambda }_1}}\right)\times {\left(1+{\lambda }_1\frac{\Phi -{\lambda }_2}{{\lambda }_3}\right)}^{-1-\frac{1}{{\lambda }_1}}$$ (2)

and is typically used to model the extremes of observation sets (e.g., maximum or minimum of repeated rounds of measurements). Mean and standard deviation of GEV are given by

$$E[\Phi ]=\mu ={\lambda }_2+\frac{{\lambda }_3}{{\lambda }_1}\left(\Gamma \left(1-{\lambda }_1\right)-1\right)$$ (3)

$$\sqrt{var[\Phi ]}=\sigma =\frac{{\lambda }_3}{{\lambda }_1}\sqrt{\Gamma \left(1-2{\lambda }_1\right)-{\left(\Gamma \left(1-{\lambda }_1\right)\right)}^2}$$ (4)

where Γ(x) is the Gamma function $\Gamma (x)={\int }_0^{\infty }{t}^{x-1}{e}^{-t}dt$. The majority of Φ observations is concentrated at low values. Empirical and model-based cumulative distribution functions (CDFs) for Φ_{C = {1,6,11}} are plotted in Fig. 3 as well as the corresponding error. Notably, 97% of Φ observations for channel 1, 99% for channel 6, and 98% for channel 11, are below 10%. Consequently, GEV is an intuitive choice to model Φ. Maximum error between empirical and GEV CDF is 4%, 1%, and 2.9% for channels 1, 6, and 11, respectively. Fitting parameters for GEV shape (λ₁), scale (λ₂), and location (λ₃) with 95% confidence intervals are reported in Table I. Findings of generally low Φ values on WLAN channels 1, 6, and 11 in a hospital environment conform with those observed in Finland [27], Italy [29], and the United States [32]. However, the use of long integration times (minutes or days) to calculate Φ could have masked the sporadic occurrences of high Φ values. In this paper, using 1-s integration time allowed Φ to be reported at a much higher time resolution than earlier studies.

Fig. 2.

Empirical and model fitted PDF of Φ_{1,6,11}. (a) Channel 1. (b) Channel 6. (c) Channel 11.

Fig. 3.

Empirical and model fitted CDF and error comparison of Φ_{1,6,11}. (a) Channel 1. (b) Channel 6. (c) Channel 11.

TABLE I.

Model Fitting Parameters With 95% Confidence Intervals

Parameter	Channel 1	Channel 6	Channel 11
Shape λ1	0.3851 (0.3842:0.386)	0.1602 (0.1597:0.1608)	0.3214 (0.3206:0.3221)
Scale λ2	0.9286 (0.9279:0.9294)	0.7914 (0.7909:0.7919)	1.1272 (1.1264:1.1281)
Location λ3	0.9907 (0.9899:0.9916)	1.2394 (1.2387:1.2401)	1.2463 (1.2453:1.2473)

Even though low Φ values are observed with high probability, high Φ values eventually occurred and can be noted by the logarithmic scale plot of the Φ histogram shown in Fig. 4. The relationship is linear for 802.11b/g and piecewise linear for 802.11n between CU and WLAN network throughput, as demonstrated in [12]. Φ observations are concentrated below Φ = 10%; therefore, if a similar CU is to be emulated in a lab environment for coexistence testing and following the findings of [12], a WLAN network can be operated at a throughput of less than 5 Mb/s when the protocols are 802.11g or 802.11n. On the other hand, rare occurrences of Φ ≈ 50% can be correlated to WLAN transmissions at about 20 Mb/s for 802.11g/n. While testing a medical device for wireless coexistence, the interchangeable CU/throughput value of the interferer with which a device can coexist successfully is quantified. This can be done by allowing the interferer to operate on its maximum possible throughput that in turn generates the maximum possible CU. The medical device then attempts to perform its wireless function. In the case of failure, the interferer’s throughput is reduced and the test is repeated until a CU/throughput value that allows the medical device to succeed is found. Consequently, Figs. 2–4 provide an estimate of the probability of observing such CU in the surveyed environment.

Fig. 4.

Histogram of Φ on WLAN channels 1, 6, and 11. Bin width is Φ = 0.25%. Note that channel 6 exhibited considerably lower maximum values compared to channels 1 and 11.

B. Temporal Distribution

In this section, we focus on the temporal aspect of Φ (i.e., the time of the day when Φ > γ was observed where γ% ∈ [0, 99] is the CU value above which Φ occurrences were noted), as illustrated in Fig. 5. Each figure corresponds to one of the three WLAN channels investigated. Using a time bin of 60 min on the x-axis, we let ζ(n, γ) be the count of occurrences of Φ > γ in time-bin n. A logarithmic scale is used on the z-axis to avoid masking high CU values by the overwhelming presence of low values.

Fig. 5.

Time distribution of Φ_{1,6,11}. Time of day n is plotted on the x-axis, CU value γ, above which Φ occurrences are counted is plotted on the y-axis and log₁₀ ζ(n, γ) is plotted on the z-axis. High CU windows are observed during daytime (on all channels) and late night hours (on channels 1 and 11). (a) Channel 1. (b) Channel 6. (c) Channel 11.

When investigating Φ > 50%, we found unique patterns for each studied channel. Maximum observed Φ values on channel 1 were logged in two main time windows: around midnight and between 3:00 and 4:00 p.m. The same high spectrum usage occurring around midnight can be noted for channel 11, in addition to an intense increase in CU between 9:00 and 11:00 a.m. As for channel 6, two high-activity windows were present between 7:00 and 8:00 a.m., as well as between 1:00 and 2:00 p.m. Increased activity around midnight can be attributed to regular data upload to a central server or routine equipment software updates, while daytime high activity windows can be related to human activity. These observations draw attention to the importance of estimating when a medical device is expected to operate in a hospital environment, as this knowledge directly affects the range of probable Φ values by reducing the sample set to those observed in the desired window.

C. Function Time Window

The time window used by a medical device to fulfill its wireless functions is of great importance in ensuring coexistence in a realistic environment. For example, a wireless-enabled sensor might report patient vital signs to a base station during a window of 5 s (i.e., the packet that holds the reported data is allowed to wait in transmission queue for 5 s before it is dropped and transmission fails). Coexistence testing can be used to determine the CU below which the device can successfully perform (i.e., maximum channel temporal occupancy that allows the device to function). To achieve this, for each test run, a CU value is set and the medical device attempts to fulfill its wireless functions. Therefore, CU integration time might be more than 1 s. In our example, CU integration time is equal to 5 s. We assume that the medical device is to be deployed in the surveyed environment after we learned empirically that Φ is distributed as f_Φ(x) = Pr[Φ = x], which is detailed in (2) with parameters from Table I. Given that the number of acquired samples per second is constant, Φ distribution over larger integration time of n seconds can be found by examining the random variable S_n

$$S_n=\Phi (1)+\Phi (2)+\cdots +\Phi (n).$$ (5)

Consequently, $\frac{S_n}{n}$ is the CU over the larger integration time of n seconds. By the law of large numbers, we know that $\frac{S_n}{n}$ approaches the mean value μ = E[Φ] as n → ∞. Mean and standard deviation for calculated and model-based Φ values are reported in Table II. To quantify the deviation from the mean for realistic integration times, we use the Chebyshev inequality

$$\text{Pr}\left[\left|\frac{S_n}{n}-\mu \right|\ge ϵ\right]\le \frac{{\sigma }^2}{n{ϵ}^2}.$$ (6)

TABLE II.

Φ Statistics

Parameter	Channel 1		Channel 6		Channel 11
	Model	Empirical	Model	Empirical	Model	Empirical
Mean, μ	2.0607	2.0903	1.8523	1.8438	2.46	2.4151
Standard Deviation, σ	3.0919	3.274	1.4864	1.3246	3.6322	2.9657

Fig. 6 shows that the probability of a deviation ϵ = 5% from the mean approaches zero using (μ, σ) values for channels 1, 6, and 11, as reported in Table II. Thus, it is evident that the longer a wireless device can wait to transmit, the more likely it is that CU will be close to the mean value. In conclusion, we note that the longer the transmission window the medical device is able to use to deliver data, the better chances it has to successfully coexist in a realistic environment. This is true because of the relatively low mean CU and its random variations seen in the actual surveyed environment, as opposed to the constant CU artificially generated in the laboratory for testing purposes, where effects on transmissions are the same regardless of the device transmission window length.

Fig. 6.

Deviation from the mean based on the Chebyshev inequality. Solid lines represent (μ, σ) obtained empirically and dashed lines for model found (μ, σ).

V. Conclusion

A long-term spectrum survey in the 2.4 GHz ISM band was conducted in a healthcare facility in the United States and a statistical distribution of WLAN CU was derived to provide support to the healthcare community to assess wireless coexistence of medical devices. The statistical distribution GEV was found to accurately fit collected CU measurements of WLAN channels 1, 6, and 11. We found that 2.4 GHz wireless spectrum, in the surveyed environment, was generally lightly used with several occurrences of high CU at various daytime and late-night hours. The results of this study can be used as an input to wireless coexistence testing or after testing to estimate the probability of wireless coexistence for similar environments. However, additional RF spectrum measurement campaigns are needed in multiple hospitals and clinics, particularly those that might have higher CU, to form a better picture of wireless patterns in healthcare facilities and the potential impact on the RF wireless coexistence of medical devices. This paper is being used to help develop a consensus standard for wireless product test methods [5] and a consensus technical information report on procedures for assessing and managing the risks associated with wireless coexistence for medical devices and systems [37].

Biographies

Mohamad Omar Al Kalaa received the Bachelor’s degree from Damascus University, Damascus, Syria, in 2008 and ME from TELECOM Bretagne, in 2012 and the M.Sc. degree from the University of Oklahoma, Norman, OK, USA, in 2014. He is currently working toward the Ph.D. degree in electrical and computer engineering at the University of Oklahoma.

He had ORISE appointment to the Research Participation Program in the Center for Devices and Radiological Health (CDRH), U.S. Food and Drug Administration. His research interests include wireless coexistence of technologies in unlicensed bands, coexistence testing methodologies, cognitive radio, applications of machine learning in wireless communication, and PHY and MAC design.

Walid Balid received the B.S. and M.Sc. degrees in electronic engineering from Aleppo University, Aleppo, Syria, in 2006 and 2011, respectively. He is currently a Graduate Research Assistant and working toward the Ph.D. degree in electrical and computer engineering at the University of Oklahoma, Norman, OK, USA.

Between 2006 and 2012, he was a principal Electronics Engineer and then an R&D Manager at AL-AWAIL, Co., Syria. In 2013, he was a Senior Research Associate at Qatar University. His main research interests include embedded system development, wireless sensor networks, and intelligent transportation system.

Hazem H. Refai received his graduate degrees from the University of Oklahoma, a master’s degree in electrical engineering in 1993 and doctorate in 1999.

He is the Williams Professor for telecommunication and networking in the OU School of Electrical and Computer Engineering Telecommunication Program, Tulsa, OK, USA. He is the Founder and the Director of the Wireless Electromagnetic Compliance and Design (WECAD) Center at OU-Tulsa. WECADs mission is to conduct basic and applied research examining medical device coexistence with various RF wireless systems and technologies, as well as validating electronic and electromagnetic compatibility. He has published more than 190 referred papers for national and international conferences and Journal articles. His fields of interest include the development of physical and medium access control layers to enhance wireless coexistence, the characterization of hospital RF environment for medical electronics, and cognitive radios and networks. He is the past IEEE ComSoc Tulsa Chapter President and served as the organizations North American Distinguished Lecturer Tour Coordinator.

Nickolas J. LaSorte received the Ph.D. degree in electrical engineering from the University of Oklahoma, Norman, OK, USA, in 2013.

He is an ORSIE Fellow in the Department of Biomedical Physics, U.S. Food and Drug Administration, Silver Spring, MD, USA.

Seth J. Seidman received the Bachelor’s and Master’s degrees in electrical engineering from the University of Maryland, College Park, MD, USA, in 2003 and 2008, respectively.

He is a Research Electrical Engineer with more than ten years’ experience at the U.S. Food and Drug Administration (FDA), Silver Spring, MD. He performs regulatory reviews, research, and has authored papers in the areas of medical device EMC and wireless coexistence. He is a U.S. Representative to International Standards Organization and International Electrotechnical Commission Joint Technical Committee 1, Subcommittee 31 on automatic identification and data capture techniques, an FDA representative to the Association for Automatic Identification and Mobility, Cochairman of the Association for the Advancement of Medical Instrumentation EMC Committee for Pacemakers and ICDs, and Vice Chair to the American National Standards Institute C63 Subcommittee 7 on Spectrum Etiquette.

Howard I. Bassen (M’77–SM’78–F’92–LF’11) received the B.S. degree in electrical engineering from the University of Maryland, College Park, MD, USA, in 1965 and the M.S. degree in management of science and technology from George Washington University, Washington, DC, USA, in 1980.

From 1972 to 1980, he was an Electronics Engineer and a Supervisory Engineer in the U.S. Food and Drug Administrations (FDA) Center for Devices and Radiological Health (CDRH). He led original research for 15 other scientists, engineers, and technicians in the measurement of electromagnetic radiation induced in the human body by radiation emitting electronic products. He was a Chief Engineer of the Microwave Research Laboratory at the Walter Reed Army Institute of Research from 1985 to 1990. He returned in 1990 to FDA/CDRH to lead the Electromagnetics and Wireless Laboratory. He has published more than 84 papers, including more than 22 journal papers and three book chapters. He holds four patents on means for measuring, transmitting, or shielding electromagnetic fields.

Mr. Bassen is a Fellow of the IEEE selected by the Engineering in Medicine and Biology Society. He chaired the ANSI C95 Subcommittee 1-Techniques, Procedures, and Instrumentation, Non-ionizing Radiation and IEEE Standards Coordinating Committee 34, Subcommittee 2 (Cellular telephone safety dosimetry techniques). He chaired the Committee on Man and Radiation in the IEEE Engineering in Medicine and Biology Society.

Jeffrey L. Silberberg received the B.S. and M.S. degrees in electrical engineering from the University of Maryland, College Park, MD, USA, in 1971 and 1976, respectively.

He has been with the FDA, College Park, MD, since December 1972. He provides electronics, electromagnetic compatibility (EMC), and standards expertise to medical device review and to the enforcement of regulatory requirements. He participates in the development of national and international standards and outreach activities to promote EMC in healthcare. He has designed, developed, and tested electronic instrumentation and systems in the enforcement of FDA regulations, and holds seven U.S. patents.

Mr. Silberberg serves as a U.S. Representative to, and the Secretary of, International Electrotechnical Commission (IEC) Subcommittee 62A, Maintenance Team 23 on EMC of Medical Electrical Equipment; as an FDA representative to American National Standards Institute (ANSI)-accredited Standards Committee on EMC, C63; and as an FDA representative to the EMC and Apnea Monitoring committees of the Association for the Advancement of Medical Instrumentation (AAMI). He served as a Principal Author of ANSI C63.18:1997 and 2014, Cochair of the AAMI EMC Committee from 1994 to 2006, Cochair of the Apnea Monitoring Committee from 1990 to 2000 and has been working on IEC 60601-1-2 since 1989. In 2013, he received the prestigious IEC 1906 Award for his contributions to improving the safety of medical electrical equipment in the electromagnetic environment.

Donald Witters received the Master of Science in Biomedical Engineering from Georgetown University in 1989, and the Bachelor of Sciences degree in Physical Sciences from the University of Maryland in 1975.

He is a Senior Biomedical Engineer/Regulatory Review Scientist in the Office of Science and Engineering Laboratories, CDRH. He has more than 40 years of experience in the areas of medical devices electromagnetic compatibility (EMC), device wireless technology, and microwave calibration. He performs laboratory research, publishes technical papers, and is an FDA Technical Expert liaison for national and international standards dealing with EMC of active medical devices such as implantable neurostimulators and pacemakers, powered wheelchairs, and hearing aids. He is the primary author of the FDA Guidance for RF Wireless Medical Devices and the new guidance for EMC Information in Regulatory Submission, and chairs the CDRH EMC and Wireless Work Group. He is also the Cochair of the Association for the Advancement of Medical Instrumentation Work Group SM/WG-06 on Wireless Medical Device Coexistence.