SYSTEMS AND METHODS FOR ENHANCING AUDIO QUALITY OF FM RECEIVERS
Systems and methods are described for enhancing the audio quality of an FM receiver. In embodiments described herein, a stop band noise signal is extracted from an L+R or L−R signal produced by an FM stereo decoder. A channel quality measure is calculated based on the stop band noise signal and is used to control whether a pop suppression technique is applied to the L+R signal. The channel quality measure and the stop band noise signal are also leveraged to perform single-channel noise suppression in the frequency domain on the L−R signal and on the L+R signal. The channel quality measure is also used to control the application of a fast fading compensation process that replaces noisy segments of the L−R and L+R signal with replacement waveforms generated via waveform extrapolation.
1 . A method for enhancing the audio quality of a Frequency Modulation (FM) receiver, comprising:
receiving an L+R signal that is output by an FM stereo decoder;
detecting noise pulses in the received L+R signal, the detecting noise pulses comprising using a plurality of noise pulse templates;
removing the detected noise pulses from the received L+R signal to produce a modified L+R signal; and
using the modified L+R signal to produce an enhanced audio signal for output by the FM receiver.
2 . The method of claim 1 , wherein each of the plurality of noise pulse templates is obtained by sub-sampling a single noise pulse at different phases.
3 . The method of claim 1 , wherein the detecting comprises:
calculating a prediction error by at least processing the received L+R signal in a short-term prediction error filter;
convolving each noise pulse template with an impulse response of the short-term prediction error filter to obtain a plurality of convolved noise pulse templates; and
comparing the convolved noise pulse templates with comparison waveforms derived from the prediction error to identify the location of candidate noise pulses in the received L+R signal.
4 . The method of claim 3 , wherein the processing the received L+R signal in the short-term prediction error filter comprises:
processing the received L+R signal in a short-term prediction error filter that includes a short-term predictor that predicts the value of a current sample as the weighted sum of a number of prior samples, wherein the prior samples are not adjacent to each other or to the current sample.
5 . The method of claim 3 , wherein the calculating the prediction error further comprises:
processing a short-term prediction error produced by the short-term prediction error filter in a long-term prediction error filter to produce a long-term prediction error.
6 . The method of claim 3 , wherein the comparing the convolved noise pulse templates with comparison waveforms derived from the prediction error comprises performing the following for each of a predetermined number of samples of the prediction error:
generating a comparison waveform corresponding to the sample;
comparing the generated comparison waveform with each convolved noise pulse template to determine a measure of dissimilarity or a measure of similarity associated with each convolved noise pulse;
identifying the convolved noise pulse template that provides the smallest measure of dissimilarity or the greatest measure of similarity; and
identifying the location of a candidate noise pulse in the received L+R signal based on the sample if the smallest measure of dissimilarity is less than a predefined dissimilarity threshold or if the greatest measure of dissimilarity is greater than a predefined similarity threshold.
7 . The method of claim 6 , wherein the generating the comparison waveform for a particular sample in the prediction error comprises:
identifying a prediction error waveform that comprises a fixed number of samples that precede the particular sample, the particular sample, and a fixed number of samples that follow the particular sample; and
normalizing and applying an offset to each of the samples of the prediction error waveform.
8 . The method of claim 2 , further comprising:
confirming that a candidate noise pulse in the received L+R signal is a noise pulse; and
subtracting a scaled version of a noise pulse template corresponding to the candidate noise pulse from the received L+R signal in response to confirming that the candidate noise pulse is a noise pulse.
9 . The method of claim 8 , wherein the confirming that the candidate noise pulse in the received L+R signal is a noise pulse comprises at least determining a measure of dissimilarity or a measure of similarity between the candidate noise pulse and a plurality of sub-vectors derived from each noise pulse template.
10 . The method of claim 8 , further comprising:
subtracting a convolved version of a noise pulse template corresponding to the candidate noise pulse from the prediction error in response to confirming that the candidate noise pulse is a noise pulse.
11 . A Frequency Modulation (FM) receiver, comprising:
an FM stereo decoder that receives an input FM radio signal and obtains at least an L+R signal therefrom; and
FM audio enhancement logic configured to:
receive the L+R signal and to detect noise pulses in the received L+R signal, the detecting of noise pulses comprising using one or more noise pulse templates that represent multiple overlapping noise pulses;
remove the detected noise pulses from the received L+R signal to produce a modified L+R signal; and
use the modified L+R signal to produce an enhanced audio signal for output by the FM receiver.
12 . A Frequency Modulation (FM) receiver for outputting an enhanced audio signal, comprising:
an FM stereo decoder that receives an input FM radio signal and obtains at least an L+R signal therefrom; and
FM audio enhancement logic configured to:
receive the L+R signal and to detect noise pulses in the received L+R signal, the detecting of noise pulses comprising:
calculating a prediction error by at least processing the received L+R signal in a short-term prediction error filter;
convolving each of a plurality of noise pulse templates with an impulse response of the short-term prediction error filter to obtain a plurality of convolved noise pulse templates; and
comparing the convolved noise pulse templates with comparison waveforms derived from the prediction error to identify the location of candidate noise pulses in the received L+R signal;
remove the detected noise pulses from the L+R signal to produce a modified L+R signal; and
use the modified L+R signal to produce an enhanced audio signal for output by the FM receiver.
13 . The FM receiver of claim 12 , wherein the FM audio enhancement logic is further configured to process the received L+R signal in the short-term prediction error filter by:
processing the received L+R signal in a short-term prediction error filter that includes a short-term predictor that predicts the value of a current sample as the weighted sum of a number of prior samples, wherein the prior samples are not adjacent to each other or to the current sample.
14 . The FM receiver of claim 12 , wherein the FM audio enhancement logic is further configured to calculate the prediction error by processing a short-term prediction error produced by the short-term prediction error filter in a long-term prediction error filter to produce a long-term prediction error.
15 . The FM receiver of claim 12 , wherein the FM audio enhancement logic is further configured to compare the convolved noise templates with comparison waveforms derived from the prediction error by performing the following for each of a predetermined number of samples of the prediction error:
generate a comparison waveform corresponding to the sample;
compare the generated comparison waveform with each convolved noise pulse template to determine a measure of dissimilarity or a measure of similarity associated with each convolved noise pulse;
identify the convolved noise pulse template that provides the smallest measure of dissimilarity or the greatest measure of similarity; and
identify the location of a candidate noise pulse in the received L+R signal based on the sample if the smallest measure of dissimilarity is less than a predefined dissimilarity threshold or if the greatest measure of dissimilarity is greater than a predefined similarity threshold.
16 . The FM receiver of claim 15 , wherein the FM audio enhancement logic is further configured to generate the comparison waveform for a particular sample in the prediction error by:
identifying a prediction error waveform that comprises a fixed number of samples that precede the particular sample, the particular sample, and a fixed number of samples that follow the particular sample; and
normalizing and applying an offset to each of the samples of the prediction error waveform.
17 . The FM receiver of claim 12 , wherein the FM audio enhancement logic is further configured to:
confirm that a candidate noise pulse in the received L+R signal is a noise pulse; and
subtract a scaled version of a noise pulse template corresponding to the candidate noise pulse from the received L+R signal in response to confirming that the candidate noise pulse is a noise pulse.
18 . The FM receiver of claim 17 , wherein the FM audio enhancement logic is further configured to confirm that the candidate noise pulse in the received L+R signal is a noise pulse by determining a measure of dissimilarity or a measure of similarity between the candidate noise pulse and a plurality of sub-vectors derived from each noise pulse template.
19 . The FM receiver of claim 17 , wherein the FM audio enhancement logic is further configured to subtract a convolved version of a noise pulse template corresponding to the candidate noise pulse from the prediction error in response to confirming that the candidate noise pulse is a noise pulse.
20 . The FM receiver of claim 17 , wherein the FM audio enhancement logic is further configured to search for a noise pulse in the received L+R signal at a location that is a predefined number of samples prior to the candidate noise pulse in response to confirming that the candidate noise pulse is a noise pulse.