Self & Ambient Noise Characterization
by Marco Flagg, 16SEP2015
All recorders and listening devices exhibit self-noise. To the extent that self-noise exceeds the strength of a signal of interest, it will obscure that signal and thus limit the detection range of the recorder. However, beyond recorder self- noise, the ambient or natural background noise in a given environment also limits the signal detection range. Thus, the fundamental goal of recorder design is to keep the self-noise level below the ambient noise level. This means that in a particular environment of interest, the signal detection range will be limited by the natural ambient conditions and not by the recorder performance.
This application note characterizes the recorder noise level in absolute terms (dB re. 1 μPa / √Hz) as well as by comparing the noise level seen by the recorder in various representative underwater environments. It evaluates noise throughout the available frequency spectrum, and identifies the specific sources of microMARS self noise. The dual purpose of the application note is to help you evaluate recorder suitability for a particular task, and to provide reference measurements against which you can check the quality of your microMARS recordings.
The microMARS hydrophone end-cap is field replaceable. Hydrophone end-caps are available with a selection of pre- amp gain with both 'flat' and sloped frequency response. They are also available with hydrophone elements designed for use at frequencies up to 33kHz (MH33) or 125kHz (MH125) respectively. For the purpose of this application note, all test recordings were done with the 33kHz hydrophone end caps in their higher sensitivity 'flat response' (MH33-2) and lower sensitivity 'flat response' (MH33-1) configurations. These two hydrophone configurations are good starting points for use of microMARS, although you may wish to consider the sloped response varieties (MH33-S1, MH33-S2) or the high frequency MH125 hydrophone once you have acquired some experience with the recorder.
Recordings were taken in six environments.
The SWFSC acoustic test tank. Located at NOAA's South West Science Fisheries Center in San Diego. it is a low-noise environment that serves to establish the recorder self-noise.
The Baltic Sea. This location, at a shallow small boat channel between small islands at Vanhaniemi near Kotka, Finland is the most quiet natural environment we have located so far. The recordings were taken during a dead calm night and the water at the location varies from about 1m to 10m. In portions of the frequency spectrum, this location was quieter than the SWFSC test tank.
The Los Padres Reservoir in Carmel Valley, CA. This is a small fresh water body selected for its potential quietness. Deployment depth about 3m in 10-15m of water. There was a slight breeze and ripples of 1-2 cm. In fact, both water sounds and distant mechanical equipment is well apparent in the recordings.
Deep water south of Catalina Island, CA. Deployed at about 100m depth in 1200m of water, 4 nautical miles from shore. It was a calm weather, and only occasional boat traffic. Distant boat traffic and sonar are readily apparent in the recordings.
Monterey Harbor, CA. This loud environment is dominated by snapping shrimp noises, pinniped vocalizations and some mechanical noises from the docks. The recorders were hung about 2m into 6m of water from the end of the Coast Guard Breakwater boat launch dock.
San Juan Island, WA. Deployments are in the Haro Straight between San Juan Island and Canada. The recordings with MH33-2 and MH33-1 were not in the same location or at the same time. MH33-2 captured killer whale vocalizations, while the MH33-1 recording shows primarily distant boating noise.
The individual recordings were spliced into a single file for MH33-1, and a second file for MH33-2. The recordings were made at a 100kHz sample rate. A 30-second sound sample is provided for each environment, and the sound samples are separated by 440Hz tones of 1 sec duration. The MH33-1 and MH33-2 sound sample files in turn are presented in RAW form (signals are dB re. FS) and in equalized form (signals are in dB re. 1 μPa / √Hz throughout the frequency range. The full scale value is 160 dB).
These sound files are published in both .WAV format for analysis, and .MP3 for listening.
MH33-1 Self Ambient Noise
The sequence of sound samples in the MH33-1 file is: SWFSC Tank, Catalina Island, Monterey Harbor, San Juan Island. Segments are separated by one second tones. Each environment is 30 seconds.
MH33-2 Self Ambient Noise
The sequence of sound samples in the MH33-2 file is: SWFSC Tank, Los Padres Reservoir, Baltic Sea, Catalina Island, Monterey Harbor, San Juan Island. Segments are separated by one second tones. Each environment is 30 seconds.
Sound Analysis Method
If you are using this data to quality-check your own microMARS recordings, use of the same data analysis method and parameter settings are essential:
For the 'raw data' spectra with response in dB re. FS
Recorder .WAV files were imported into Audacity, a free software, to compute the frequency spectrum. Each plot corresponds to the full 30-second sound sample, i.e. the results are averaged over that period (Long Term Spectral Averages). For the full bandwidth analysis to 50 kHz, the settings as in the screen dump below were used: FFT size 8192, and Hanning window. These settings are compatible with the microMARS quality control tests, and thus the data in this application note can be compared to the QC report for any particular recorder. The FFT size of 8192 also produces a reasonable frequency bin width of 12.2Hz (plotting with narrower bins boosts noise peaks but reduces the noise floor, while wider bins result in attenuated noise peaks but higher noise floors as peak energy is averaged over a wider frequency range). For the low frequency analysis to 100 Hz, a FFT size of 65536 was used to obtain a good frequency bin resolution of 1.52 Hz.
Figure 1: Audacity spectrum analysis settings for frequency analysis up to 50 kHz. A FFT size of 65536 is used for the low frequency analysis to 100Hz
The exported Audacity spectrum data was imported into Excel to adjust for the industry standard noise bandwidth of 1Hz, and to make 0 dB FS correspond to a full scale square wave or a full scale DC or instantaneous value (as opposed to a full scale sine wave, which will register at -3 dB): Corrected dB = Audacity spectrum value - 10 * LOG10 (FFT Bin Bandwidth) - 20 * LOG10 (√2). The corrected value is what appears in the spectrum graphs in this app-note.
Cool Edit, a software also available free of charge, was used to produce a waterfall spectrum display, providing frequency vs. sound power over time.
For equalized spectra with response in dB re. 1 μPa / √Hz
After import into Audacity, the sound sample data is equalized using the equalization curves provided by Desert Star for the MH33-1 and MH33-2 hydrophones. These equalization curves are based on a frequency response modeling of the hydrophone element, the coupling of the hydrophone to the pre-amplifier, the pre-amplifier response and the final amplifier response. The equalization curves are published with this application note.
Equalization with the standard curves results in a full scale value of 180 dB re. 1 μPa. Since the analysis focused on weaker signals and to improve audibility of the sound files, a 20 dB gain was commanded, resulting in a full scale value of 160 dB re. 1 μPa.
The spectrum analysis using the same settings as for the 'raw data' analysis followed.
The spectrum data was exported from Audacity, imported into Excel and corrected as follows: Corrected dB = 160 dB + Audacity spectrum value - 10 * LOG10 (FFT Bin Bandwidth) - 20 * LOG10 (√2). The 160 dB offset reflects the acoustic pressure corresponding to a full scale value.
Raw (non-equalized & normalized) Noise Measurements
The following plots shows the noise observed by the recorders in the various test environments, expressed in dB re. FS.
The major observations are:
The noise level observed in most of the environments was higher than that observed in the SWFSC tank (red line). That means the ambient noise, not the recorder self noise, is the detection range limiting factor in these environments. An exception was the sound sample collected in the most quiet environment, the Baltic Sea shallow location during a calm night. Here the noise observed in some components of the spectrum was lower than in the SWFSC tank, suggesting that even in the test tank, the recorder still picks up ambient noise.
The noise level in the SWFSC tank for the most part is in the -90 dB or lower, meaning that the dynamic range of the converter (the room between the signal clipping level and the noise floor) is preserved. Note that in some cases the noise floor is below the -96.3 dB quantization limit for a 16-bit A/D converter. This is possible because the noise in the graphs is expressed per √Hz rather than the noise power over the entire spectrum.
Figure 2: Raw observations for MH33-2 and MH33-1 in various test environments. The observed ambient noise for most environments is higher than the noise observed in the assumed quiet SWFSC tank (red line). Thus, recorder performance will generally be ambient noise limited, not self-noise limited.
Figure 3A: Waterfall spectrum display for MH33-2. The horizontal axis is time and vertical axis is frequency. Six recording segments are shown corresponding to (from left) SWFSC Test Tank, Los Padres Reservoir, Baltic Sea, Catalina Island deep water, Monterey Harbor and San Juan Island respectively. Lighter streaks show detected signals. Only the SWFSC test tank and the Baltic sea at night locations appear without obvious ambient signal power. All other environments had ambient components clearly above the recorder self noise level.
Figure 3B: Waterfall spectrum display for MH33-1. Environments (from left) are SWFSC test tank, Catalina Island deep water, Monterey Harbor and San Juan Island.
Equalized and Normalized Noise Measurements
The following plots shows the noise observed by the recorders in the various test environments, expressed in dB re. 1 μPa / √Hz The major observations are:
MH33-2 provides a lower noise floor in the SWFSC tank. So, it's use is beneficial for weaker signals.
MH33-2 also provides a higher differential between the self noise floor and ambient environment, also implying its benefit for weak signal detection.
Figure 4: Equalized and normalized observations forMH33-2 and MH33-1 in various test environments. The observed ambient noise for most environments is higher than the noise observed in the assumed quiet SWFSC tank (red line). Thus, recorder performance will generally be ambient noise limited, not self-noise limited.
Low Frequency Noise Measurements
These measurements provide higher resolution in the 1-100 Hz frequency range for the same set of recordings by using a larger FFT (smaller frequency bin size of 1.52 Hz). Most environments shows higher ambient noise levels than the noise observed in the SWFSC tank. However, the Baltic Sea location shows a lower noise floor than the SWFSC tank, suggesting some low frequency noise in the tank and a recorder self-noise floor in this frequency range at or below the signal observed in the Baltic Sea.
Figure 5A: Low frequency measurements forMH33-2, in raw form.
Figure 5B: Low frequency measurements for MH33-2, in equalized and normalized form.
Characterization of the microMARS recorder self-noise
The individual components of the recorder self-noise can be tied to specific design characteristics of the recorder. As these noise components are predictable, they can be filtered out through notch filters or other appropriate filtering if necessary. Comparing figure 6 below to figure 4 shows that any particular self-noise spike will only manifest if the operating environment is sufficiently quiet.
Knowledge of the recorder's characteristic noise profile is also useful when trying to determining if a given noise line observed in a test recording is recorder self-noise or induced noise such as EMI or RFI.
Figure 6: Self-Noise components of microMARS with MH33-2
Bank Switching Noise: microMARS alternately stores sound on pairs of two cards, storing 256 samples on bank A, followed by 256 samples on Bank B before returning to Bank A. This results in a bank switching noise at 1/512th of the sampling frequency, or 195 Hz for a 100kHz sampling rate.
Bank switching harmonics: Bank switching noise can be thought of mostly as a square wave, with equal Bank A and Bank B components. The first harmonic of a square wave is at three times the switching frequency or 585 Hz. Weaker harmonics manifest at higher frequency multipliers.
Crystal oscillator noise: microMARS timing is controlled by a crystal oscillator operating at 32.768kHz. Thus, a noise line manifests at that frequency.
Suspected facility noise: A broad signal hump manifested in the SWFSC test tank between about 20Hz and 200Hz. Since this hump was not seen in a recording at the quiet Baltic Sea site, it is suspected to be facility noise.
Amplifier noise floor: Pre amplifier noise manifests from amplifier input leakage currents (input referred noise)that work in combination with the hydrophone element impedance to generate a voltage across the amplifier input terminals. In combination with the sensitivity of the hydrophone element (specified as the open circuit voltage response), this in turn defines the acoustic noise floor of the hydrophone/pre-amp design. The final amplifier stage contributes noise in a similar fashion although it tends to be a smaller contributor due to the lower amplification of noise originating at that stage.
Hydrophone resonance frequency: Hydrophones are normally operated well below their resonance frequency to stay in the range of flat frequency response. At resonance, hydrophone impedance declines, and so self noise will appear diminished. However, acoustic sensitivity also increases, and therefore any ambient noise at the frequency will manifest as a noise band. The microMARS MH33 hydrophone is specified for use up to 33kHz, and has a resonance frequency of 48 kHz.
On account of amplifier noise and its interaction with the hydrophone element, the overall noise floor of microMARS is subject to the configuration of the hydrophone pre-amplifier. In general, a less sensitive configuration like MH33-1 will accommodate larger signals without clipping but also exhibit a higher noise floor. Best sensitivity and the lowest noise floor is obtained with end-caps that have a sloping response, i.e. exhibit a higher voltage gain at higher frequencies. The sloped response is equalized as an early step in signal processing, and the result is a file that can accept loud low- frequency signals yet is also very sensitive at high frequencies. As a result, the dynamic range of MH33-S2 for example is 150 dB, extending from about a 22dB noise floor at 33kHz to a 172 dB clipping point at 100Hz.
Figure 7: Noise floor measurements for the four standard configurations of the MH-33 hydrophone endcap