Tweet
The Time Domain Adaptive Filter (TDAF)
(Forensics Version Only – Advanced Filters)
This Time Domain Adaptive Filter (TDAF) is used largely in audio applications where the ambient noise environment is constantly changing and the filter coefficients must automatically adapt to maintain good intelligibility of an audio signal. This is the Time Domain based “older sister” to the forensics Adaptive Frequency Domain Filter (Forensics AFDF) found inside the Diamond Cut Continuous Noise Filter (one of the CNF Modes).
The Time Domain Adaptive Filter
The Adaptive filter adjusts itself to remove a modeled signal representing the unwanted time domain waveform while preserving the target signal. It uses an advanced form of an adaptive least mean squared algorithm to provide continuous adaption of the filter coefficients. It works best with a reference signal (in other words, a stereo or binaural source) containing only the noise to be rejected. This second channel or reference signal can be obtained from a second surveillance track with its microphone located near the noise source in the room such as a Jukebox or a television set. However, it can also use its own signal as a reference in conjunction with the time delay function, which is provided for monophonic situations. Additionally, the adaptive filter provides either the main processed signal or a keep-residue mode output signal for rejecting a wide variety of different types of noise sources. Sometimes it will be found that the “keep residue” mode signal is more useful than the main output signal. Trial and error is sometimes the best way to determine the best mode to use. The following controls are included with the adaptive filter:
Important Note:
Because of the wide range of convergence values allowed in the Adaptive Filter, certain audio signals may cause the filter to become unstable and cut out. In these cases, try changing the Convergence parameter until the audio is restored. Usually lower convergence settings are more stable, but adapt more slowly to changes in the audio signal. Lengthening the filter (Samples) may also increase its stability.
(Forensics Version Only – Advanced Filters)
This Time Domain Adaptive Filter (TDAF) is used largely in audio applications where the ambient noise environment is constantly changing and the filter coefficients must automatically adapt to maintain good intelligibility of an audio signal. This is the Time Domain based “older sister” to the forensics Adaptive Frequency Domain Filter (Forensics AFDF) found inside the Diamond Cut Continuous Noise Filter (one of the CNF Modes).
The Time Domain Adaptive Filter
The Adaptive filter adjusts itself to remove a modeled signal representing the unwanted time domain waveform while preserving the target signal. It uses an advanced form of an adaptive least mean squared algorithm to provide continuous adaption of the filter coefficients. It works best with a reference signal (in other words, a stereo or binaural source) containing only the noise to be rejected. This second channel or reference signal can be obtained from a second surveillance track with its microphone located near the noise source in the room such as a Jukebox or a television set. However, it can also use its own signal as a reference in conjunction with the time delay function, which is provided for monophonic situations. Additionally, the adaptive filter provides either the main processed signal or a keep-residue mode output signal for rejecting a wide variety of different types of noise sources. Sometimes it will be found that the “keep residue” mode signal is more useful than the main output signal. Trial and error is sometimes the best way to determine the best mode to use. The following controls are included with the adaptive filter:
- Convergence (Adaptation Speed) - Slower adaptation speeds produce better noise rejection for stationary noise (not changing), while faster adaptation speeds produce better adaptation response in quickly varying ambient noise conditions.
- Filter Length (Samples) - The larger this number the more signal inflection points can be modeled in the time domain signal in order to be rejected or maintained. Generally speaking, a “sweet spot” is often found between 10 and 100 samples. Very long filter samples can be very demanding upon your processor’s resources.
- Reference Signal - This is the signal to be compared to when a stereo recording is available. (Choose the one that contains the reference signal to be used.)
- [*=3]Right Channel [*=3]Left Channel [*=3]Time Delay (when there is no reference signal)
- Time Delay (Samples) - For use in time delay reference mode only when no reference signal is available. This essentially allows a delayed representation of the signal being adapted to be its own reference.
- Adapt / Freeze button - Selects adapt or freeze coefficients mode of operation. Usually this is activated in Adapt mode for the first 5 seconds of noise. In some cases where the sound ambient is constantly changing, one may choose not to “freeze” this filter.
- Keep Residue button - Allows the operator to use the error signal rather than the output signal from the Adaptive filter. This feature is useful for attenuating continuous loud or varying tones (like a siren) that may be masking a Forensics recording.
- Threshold - This control sets the level above which the system re-initializes itself based on the applied signal amplitude. Generally, you will find that settings somewhere between –20 dB and –40 dB are useful values. If this control is set to 0 dB, it will not produce any output because it will be continuously re-initializing itself.
- Threshold LED Indicator - Just to the right of the threshold control is a Green LED indicator. When this indicator is illuminated, the adaptive filter is active and adapting to the signal that is present. If the LED is not illuminated, move the Threshold Control downwards until the Green LED flashes or illuminates in order for proper filter operation to occur.
- Graphical Display - You can choose between two graphical display modes. One mode displays the Amplitude of the Filter Coefficients on the vertical axis (on a normalized basis) vs. the Index of the Filter Coefficients (which is an indicator of how many filter taps or filter length which are being used). The second mode plots the frequency response (relative amplitude vs. frequency) being produced by the adaptive filter on a time-updated basis.
- Adaptive Normalize - This button, when checked, places the system into a “Normalized Least Mean Squared” mode, which is useful when attempting to clean up material with widely varying signal amplitude components. This is the preferred default mode of operation.
Important Note:
Because of the wide range of convergence values allowed in the Adaptive Filter, certain audio signals may cause the filter to become unstable and cut out. In these cases, try changing the Convergence parameter until the audio is restored. Usually lower convergence settings are more stable, but adapt more slowly to changes in the audio signal. Lengthening the filter (Samples) may also increase its stability.