Field Evaluation of Reflected Noise from a Single Noise Barrier—Phase 1 (2016)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

7 C H A P T E R 2 Research Approach As noted in Chapter 1, the objectives of this research are to: 1. Determine the spectral noise level characteristics of the overall noise in the presence of a single reflective noise barrier for positions on the opposite side of a roadway through the collection of field measurements from diverse sites, and 2. Summarize and analyze the implications of the research results for purposes of understanding the actual and perceived effects of reflected noise. Based on its understanding of these objectives and the nature of the problem, the research team investigated the changes in the broadband A-weighted sound levels and unweighted sound pressure levels and individual 1/3 octave band sound pressure levels between No Barrier and adjacent Barrier sites. In addition, the following components were examined: (1) the spectral time signature of the signals of individual passbys with and without the far-side barrier; and (2) the difference in psychoacoustic sound quality metrics. Research Tasks This research consisted of six tasks. Each task is described briefly below. Task 1. Kick-off teleconference meeting and amplified work plan development Task 2. Completion of literature review, which is presented in Appendix A Task 3. Development of study location selection criteria, identification of recommended study locations, and data collection, processing and analysis protocols Task 4. Measurements and analysis for the first two study locations Task 5. Measurements and analysis for the remaining locations and summary and analysis of implications of all research results Task 6. Preparation and delivery of draft final and final report and PowerPoint™ presentation Data Collection Protocol Three types of data analysis were used in this study: 1. FHWA Method (based on the Indirect Measured procedure in Chapter 6 of FHWA’s Measurement of Highway-Related Noise (Ref. 3)), where simultaneous A-weighted and 1/3 octave band measurements are made at the Barrier site and an “equivalent” No Barrier site 2. Spectrograms 3. Psychoacoustic metrics A single data collection process yielded the data for use in all three analyses, and is discussed first. Then the data processing steps are outlined. Finally, the data analysis methodology for each type of data

8 is discussed. In addition to the sound level data, the data collection for the FHWA Method included recording of calibrated .wav audio files at each microphone position. These files were used in this study for processing and analysis of the spectrograms and the psychoacoustic metrics. The FHWA Method calls for equivalence of site geometry, noise sources, and meteorological parameters. As a result, the locations were chosen such that the No Barrier and Barrier sites were adjacent. Since the sound level and frequency spectrum produced by traffic are affected by the pavement and the roadway grade; these factors were considered in site selection and were required to be equivalent in the No Barrier and Barrier locations. Two other source characteristics that can affect the sound level and frequency spectrum are the volume and speed of the traffic. Because one goal in site selection was to avoid interchanges or intersections between the Barrier and No Barrier sites and because the Barrier and No Barrier measurements were to be made simultaneously, any potential variations in traffic volume and speed between the two sites was minimized. However, traffic volumes and speeds vary over time. Likewise, experience indicates that meteorological conditions, particularly the wind speed and direction, can change even over relatively short periods of time. Therefore, time periods were grouped under equivalent source and meteorological conditions. Measurement of Highway-Related Noise (Ref. 3) recommends a minimum of three “measurement repetitions” per site (with a preferred number of six repetitions) under equivalent source and meteorological conditions. The challenge is that it can be very difficult and time-consuming while in the field collecting data to demonstrate source equivalence and meteorological equivalence real-time. As a result, the field protocol was to: • Collect four hours of data at each site, in one-second logging intervals, to be aggregated into one-minute periods; • Video-record the traffic and measure speeds with a laser speed gun; and • Collect wind speed and direction and temperature data at two heights, to be averaged in one- minute periods. Then, as part of the data processing protocol, the four hours of data were divided into one-minute periods of equivalent source and meteorological conditions at the No Barrier and Barrier sites. Each period represents one “measurement repetition” for a unique combination of equivalent source and meteorological conditions. A method of determining source equivalency between periods—based on the reference microphone sound levels and the average speeds by direction of travel—was found to work very well and was adopted for the study. Basing the source equivalence determination on the measured reference microphone data allowed the maximum tolerance between two periods to be 0.3 dB or less across all of the locations. The wind data were processed to determine a vector wind speed (component of wind speed perpendicular to the road) and corresponding wind class (Upwind, Calm, or Downwind) for each period. The temperature data were also processed to determine the corresponding temperature gradient class (Lapse, Neutral, or Inversion) for each period. These data were merged with the source data and sorted to determine the combined equivalent source and meteorological data period repetitions. Equipment and Sound Level Descriptors A standardized data collection equipment package was used at all of the sites, consisting of: • Six Type I sound level analyzers with 1/3 octave band measurement capabilities with data loggers and audio recorders;

9 • Meteorological data collection station with precision temperature sensors and precision quality anemometers capable of measuring wind speed in three dimensions at two heights (5 ft and 15 ft) above the ground; • Video camera and laser speed gun; and • Accessories including a 94-dBZ calibrator, extension cables, windscreens, microphone tripods and stands, extension poles, and guy wires. Depending on site characteristics and goals at any of the measurement locations, the community microphones could be located at two different distances from the road or at two different heights at the same distance. Regarding the naming of the microphones, when two microphones were at different heights at the same distance, at the Barrier site, the lower microphone was named BarCom03 and the upper one BarCom04 (Figure 2). At the No Barrier site, NoBarCom05 was the lower microphone and NoBarCom06 was the upper microphone. When the community microphones were at different distances, BarCom04 and NoBarCom06 were the distant microphones for the Barrier and No Barrier sites, respectively. The reference microphones on the barrier side of the road were named BarRef01 for the one adjacent to the Barrier and NoBarRef02 for the No Barrier site. Figure 2. Typical microphone positions. Measurements were made in terms of the equivalent sound level, Leq, for the broadband (overall) A-weighted sound level and unweighted sound pressure level and individual 1/3 octave band sound

10 pressure levels. One-second broadband A-weighted levels and unweighted sound pressure levels and 1/3- octave band unweighted sound pressure levels between 12.5 Hz and 20,000 Hz were saved, and were later processed into one-minute intervals, and the audio signal was recorded. Analysis of the initial data led to a decision to only present data in the 20 Hz to 10 kHz bands to eliminate the distraction of data irrelevant to the study and undue influence on the broadband unweighted sound pressure levels and A-weighted sound levels. These broadband levels were recomputed after the very low and high bands were deleted. The main reason for including both the A-weighted and unweighted data was to see if there was a difference in the two, which might provide an initial indication of frequency-specific effects. For example, higher unweighted levels might point toward substantial lower frequency components of the received sound. On the other hand, if the A-weighted level was close to or higher than the unweighted levels, there could be important contributions in the 1,000 Hz to 4,000 Hz bands. Studying only the A- weighted levels could disguise lower frequency contributions. The Barrier and No Barrier spectra were studied in terms of unweighted sound pressure levels to give a true picture of the spectra and the unweighted levels were then the basis for the Barrier/No Barrier 1/3 octave band level difference comparisons described in the FHWA Method Data Analysis Protocol section below. Statistical exceedance descriptors, specifically L1, L5, L10, L33, L50, L90 and L99, were computed for the five-minute periods based on the one-second data. These “Ln” descriptors were used in determining the sound level range in a sample period and in diagnosing data on individual passbys and the possible sustaining of the background level due to sound reflections off the barrier. Meteorological Data At each location, the meteorological station was set up in an open area near the No Barrier site. The wind data were used to determine a vector wind speed in the direction from the roadway to the microphones in order to be able to determine the appropriate wind class for the measurements. Temperature data at 5 ft and 15 ft above ground level were used to determine the appropriate gradient class for the measurements. Measurement Procedures Prior to going into the field, the measurements were planned in detail using a field review report as a guide. On the measurement day, the team set up and calibrated the sound level analyzers and deployed the microphones via extension cables on tripods or guyed towers. In addition to calibration, the electronic noise floor of the entire acoustic instrumentation system was established. Approximately four hours of simultaneous data were then collected at all of the microphones. The meteorological data were saved as one-second wind speed and direction and temperature for later processing into one-minute averages, time-synched to the sound level data. Each noise measurement person kept field logs of events, with the time of occurrence of vehicles of interest (typically heavy trucks) and any unrepresentative sounds or events that might affect the one- minute measurements. These latter events were studied for possible elimination of the one-minute data intervals from the analysis. Samples of vehicles speeds were stored in a file in the laser gun and were also recorded manually on data sheets to identify the vehicle type. Speeds varied by lane, as expected. For consistency, for roads with more than two lanes in each direction, the majority of speeds were measured in the second lane from the outside. Additional samples of speeds were made in the other lanes to the extent possible. At the end of the measurements, the calibration was checked for sound level analyzers, with the audio of the calibration tone recorded. All of the data were then downloaded onto personal computers, with a common file-naming convention for all of the files. Vehicle classification counts (automobiles, medium trucks, heavy trucks, buses and motorcycles) were made from the video in one-minute intervals that matched the sound level measurement intervals. The speed data were entered into the speed spreadsheet by ID number and vehicle type. The speeds were

11 adjusted to account for the angle of speed shooting off of head-on. As needed, the speed samples were time-adjusted forward or backward to represent the time of passage from the shooting point to a point midway between the Barrier and No Barrier sites. FHWA Method Data Processing Data processing for the FHWA Method involved three major steps: creation of data spreadsheets, elimination of time periods with unrepresentative events that affected the measured sound levels, and identification of equivalent time periods in terms of meteorological class and traffic parameters. First, the sound level and meteorological data were processed into a single standardized spreadsheet format for both the “raw” one-second data and one-minute interval (or “period”) data averaged from the one-second data. The sound level data included the A-weighted sound level and unweighted sound pressure level plus the 1/3 octave band sound pressure levels. The meteorological data included the average wind speed, wind direction, temperature, and relative humidity at each sensor (high and low heights of 15 ft and 5 ft). The vector component of the average wind velocity in a perpendicular line from the highway to the reference microphone was computed for each period, as well as the temperature gradient. Each period was classified by wind class based on Table 1 (which is Table 3 from Measurement of Highway-Related Noise). Winds outside these conditions (for vector components over ± 11 mph) were put into a class called Invalid-wind. Table 1. Classes of wind conditions. Wind Class Vector Component of Wind Velocity Upwind -2.2 to -11 mph Calm -2.2 to +2.2 mph Downwind +2.2 to +11 mph Each period was also classified by temperature gradient class, per Table 2. These classes are based on data collected by ATS from the Arizona Transportation Research Project (Ref. 4) several years ago. The Neutral conditions are based on the graphs presented in that report. Then, based on the wind class and temperature gradient class, each one-minute period of sound level data was put into one of ten meteorological classes (Upwind Lapse, Calm Lapse, Downwind Lapse, Upwind Neutral, Calm Neutral, Downwind Neutral, Upwind Inversion, Calm Inversion, Downwind Inversion, and Invalid-wind). Table 2. Classes of temperature gradients. Temperature Gradient Class Gradient: (Temp_upper – Temp_lower) Divided by Vertical Distance between Sensors Inversion positive > 0.1 Neutral -0.1 to 0.1 Lapse negative < -0.1 Based on the field notes, the data were screened for any potentially bad or unrepresentative events at each microphone position (e.g., loud non-traffic noises, periods of stopped traffic flow, etc.). As needed, the one-second data and one-minute averaged data were reviewed to see if the events affected the levels. The next step was to determine five-minute periods that were equivalent to each other for inclusion in a measurement repetition “group.” First, five-minute running averages of the vector wind component were computed for each minute of the four-hour block (excluding those five-minute periods that had one or

12 more bad one-minute periods). “Five-minute running averages” means that each consecutive minute is the starting minute of a five-minute period including its data and the data in the next four minutes. For example, 12:01 to 12:06, 12:02 to 12:07, and 12:03 to 12:08 would be three consecutive running five- minute periods. The use of five-minute running averages gives more flexibility when trying to determine periods that have equivalent sources and meteorological conditions. Each five-minute period was assigned to a meteorological class, based on a requirement that at least three of the five minutes be in the same class. All of the five-minute periods in the same meteorological class that were not overlapping in time with each other were then tested for traffic equivalence. An example of overlapping periods would be 13:45 to 13:50 and 13:47 to 13:52, whereas 13:45 to 13:50 and 13:50 to 13:55 would be non-overlapping. Finally, traffic equivalence was determined. FHWA Method Data Analysis Protocol After the equivalent five-minute periods were determined for the different meteorological classes and traffic conditions, each grouping of non-overlapping equivalent periods was used to compute the sound level increases between the Barrier and No Barrier microphones. The data analysis procedure in Measurement of Highway-Related Noise (Ref. 3) was used, with some adjustment. The first step was to determine any needed calibration adjustments prior to data analysis. The procedure in Section 6.6.3 of Measurement of Highway-Related Noise was adapted for levels measured opposite the noise barrier rather than behind the barrier, and for analysis in 1/3 octave bands. Details are in Appendix B. The sound level changes between No Barrier and Barrier sites were determined for the different groups of equivalent five-minute periods. The mean broadband A-weighted and unweighted level changes and individual 1/3 octave band sound pressure level changes were computed for each group by arithmetically averaging the differences from the individual five-minute periods. A standard deviation was computed for each sound level increase and the results were plotted. The average differences by frequency band were then computed for all equivalent five-minute periods that were analyzed within a meteorological class occurring at each location. Differences of these averages differences were also computed to allow study of the possible effect of meteorological class on the results. After study of the average differences for each equivalent group for each meteorological class, it was decided that the average differences by meteorological class represented the individual groupings’ difference quite well. Thus, these latter average differences by meteorological class are presented in this report. All of the graphs and tables of the average differences by the groups of equivalent five-minute periods are in the spreadsheets in the project record. Finally, the differences of the Ln values for each five-minute period were computed and analyzed. Spectrogram Data Processing and Analysis Protocol A spectrogram analysis allows an examination of spectral (frequency) content over time, whether it is a specified time block (e.g., 5 minutes) or a vehicle passby event. As noted above, the research team screened the master raw data spreadsheet files from the measured sound level data files and identified bad or invalid data periods. Clean data blocks were identified for the spectrogram analysis; the length of these data blocks varies among the sites, based on how often the intrusive noise events occurred and also whether or not it was desirable to examine the same five-minute data blocks as were examined with the standard analysis. Example data blocks for each site are shown in this report. In addition, vehicle passby events were identified for investigation. The vehicle passby events were first identified using the site logs, where potential isolated events were noted. Multiple events were examined for each site and only ones that could be clearly identified at both the Barrier and No Barrier sites were retained. Example vehicle passby events for each site are shown in this report.

13 The audio .wav files were processed and examined in 1/3 octave bands in 1/8-second intervals. The data were then displayed using spectrogram-type graphs. In-house MATLAB code allows for the spectrogram processing and the flexibility to compare selected time blocks of data among different microphones at each site. For the data blocks and vehicle passby events, pairs of microphones were compared, where each pair consisted of one microphone at the Barrier site and an equivalent microphone at the No Barrier site. For all sites, spectrogram data were examined for the community microphone pair BarCom03/NoBarCom05 and the community microphone pair BarCom04/NoBarCom06. The spectrograms for these two pairs show the effect of the barrier noise reflected back across the highway to communities opposite a noise barrier. At some of the sites, the reference microphones were strategically placed between the road and the barrier to capture barrier reflections on the barrier side of the highway. In these cases, microphones BarRef01 and NoBarRef02 were compared to show the effect of the barrier-reflected noise close to the reflecting surface. Upon examination of the spectrograms, spectral shapes and values were compared for the equivalent microphone pairs, and trends were noted. Where results were similar between microphone pairs, only one pair is presented. Psychoacoustics Processing and Analysis Protocol The audio recordings made by the measurement team for each location were calibrated, post- processed and analyzed for the psychoacoustic parameters of interest. In some cases, audio filters were applied, especially at higher frequencies, to remove electronic artifacts prior to post-processing. The Psychoacoustics Processing and Analysis Protocol in Appendix A outlines the background and details behind the psychoacoustic processing selected for this study. Summarizing the key points of that discussion: 1. The processing is based on a subset of industry-accepted psychoacoustic metrics. 2. Those metrics are combined parametrically to estimate potential annoyance based on three algorithms from extant sound quality literature. 3. The metrics are computed as functions of time, in one-minute intervals, for each recorded microphone signal. 4. Descriptive statistics are derived from each metric’s time series to examine trends and differences between locations. The psychoacoustic metrics applied to the audio recordings include: • Loudness in sones; • Sharpness in acums; • Roughness in aspers; • Fluctuation strength in vacils; Not all of the psychoacoustic metrics have an internationally standardized algorithm. Therefore, among the commercially available software offerings, there is not a fixed calculation associated with each metric. As a result, all of the psychoacoustic analyses were completed using a software package developed specifically for this project by Nelson Acoustical Engineering, Inc. The algorithms encoded into the software comply with those International Standards currently available; they make use of widely accepted expressions for those metrics that are not standardized. The formulae applied in the psychoacoustic analysis software are detailed in Appendix B. It is also important to take into account the time-varying Loudness created by traffic noise, since a barrier will create additional temporally varying sounds due to reflections. The time-varying Loudness created by traffic noise in these cases can cause interaural time differences (ITDs), interaural intensity differences (IIDs), and spectral changes that can create localization errors among other issues. To help

14 account for the addition of the temporally varying sounds statistical values of Loudness have been calculated. While these individual psychoacoustic metrics are useful for comparing sounds of widely different natures, they do not necessarily indicate whether an individual might be annoyed by a given sound, nor to what extent. Several annoyance scales have been suggested in the literature; each is based on regressions over listener preference trials, making use weighted combinations of the psychoacoustic parameters. The metrics demonstrated in the current work are detailed in the Psychoacoustics Processing and Analysis Protocol in Appendix B. They include: • Unbiased Annoyance (UBA), using Loudness exceeded 10 percent of the time (N10), Mean Sharpness (Sm), and Mean Fluctuation Strength (Fm) • Psychoacoustic Annoyance (PA), using Loudness exceeded 5 percent of the time (N5), Mean Sharpness (Sm), and Mean Fluctuation Strength (Fm), and Mean Roughness (Rm) • Category Scale of Annoyance (CSA), using Loudness exceeded 5 percent of the time (N5), Sharpness exceeded 50 percent of the time (S50), and Mean Roughness (Rm) For each site, for the corresponding community microphone locations in the presence and absence of the barrier, respectively, the annoyance metrics computed as a function of time are paired for comparison. Time intervals are the same as those used in the spectral analysis (one minute); this time interval is sufficiently long that short-term events, such as truck passbys, are captured at both monitoring locations. Because the annoyance metrics UBA and PA are sensitive to Sharpness, the high-frequency content in the recordings is crucial. Some of the digital audio recordings used in this work had some high-frequency contamination due to electronic noise. Audio filtering was applied to the contaminated recordings to reduce this effect. Therefore, the annoyance metrics became sensitive to the applied filters. As a result, the Briley Parkway data were not analyzed, and the annoyance metrics computed for I-24 may be slightly biased by their filtering. The I-90 and SR-71 recordings had some contamination, but it was at lower magnitude: filtering was mild in this case, and the annoyance metrics are likely internally consistent for those sites. The recordings from MD-5 were free from contamination. However, the presence of biologics (tree frogs) in the night recordings required additional filtering. In this case, however, the same filter was applied to all of the recordings, so the resulting metrics are still comparable.