This document describes PBFDAFAEC.EXE, the command line Acoustic Echo Canceller (AEC) based on the Partitioned Block Frequency Domain Adaptive Filter (PBFDAF) for 32-bit Windows operating systems. Like BFDAFAEC, PBFDAFAEC lets you change the adaptive filter parameters, run the AEC on two audio input files, namely the Far End Speech (FES), and Near End Speech (microphone) signals, and writes the echo free signal to an output file. PBFDAFAEC will also print the processing time, and you can instruct it to save the filter coefficients for later examination. PBFDAFAEC is meant to serve as a demonstrator for the performance of the PBFDAF algorithm and but can also be used as a design or a teaching tool.

The software described in this document is protected by copyright laws and international treaty provisions. DSP ALGORITHMS grants you the right to use this software under the terms and conditions stated in the accompanying file "license.txt". Please make sure that you carefully read and fully understand the license agreement before installing or using this software.

This demo version of the acoustic echo canceller described here does not include many functions available in DSP ALGORITHMS' commercial echo cancellers such as double-talk detector, far end speech detector, near end speech detector, control unit, automatic gain control, non-linear processor, or noise canceller. It implements the adaptive filter block only. Future versions may include more features as appropriate for the purpose of demonstrating the capabilities of the full version.

PBFDAFAEC prints the time it takes to process the complete input files. This processing time is estimated using the system time and therefore will be accurate enough for long input files. You can compare the processing time in seconds with the duration of the input files in seconds to see whether your machine will meet the real-time requirements for the specific AEC parameters you have chosen.

Guavec, NLMSAEC, BFDAFAEC, and PBFDAF were all originally intended as design tools to help DSP ALGORITHMS' customers in understanding and integrating the AEC software into their products. The software however proved very useful in teaching engineers the basics of adaptive filters in general, the PBFDAF algorithm in particular, and the application of adaptive filters in acoustic echo cancellers.

Acoustic echo is inevitable whenever a loudspeaker is placed near a microphone in a full-duplex communication application. This is the case in speaker-phones, audio and video conferencing, desktop communication, and many other communication scenarios. Especially hands-free mobile communication kits for cars are becoming increasingly important due to safety regulations introduced in more and more countries. In all those and similar communication scenarios, the voice from the loudspeaker (Far End Speech, FES) is inevitably picked by the microphone (NES) and transmitted back to the remote speaker as shown in Fig. 1. This makes the remote speaker hear her own voice distorted and delayed by the communication channel, which is known as echo. The longer the channel delay, the more annoying the echo becomes until it makes natural conversation impossible and decreases the perceived quality of the communication service. It is therefore absolutely necessary to avoid transmitting back the echo picked by the microphone.

Modern full-duplex communication systems make use of an acoustic echo canceller (AEC) to prevent the echo from being transmitted back to the channel. The AEC is employed in each terminal, and has completely different requirements than the Network Echo Canceller employed by the telephone network provider to eliminate the electric echo. The AEC basically estimates the echo and subtracts the estimated echo from the microphone signal as shown in Fig.1. The resulting signal (RESidual) is transmitted to the far end speaker through the communication channel.

PBFDAFAEC.EXE accepts command line parameters to allow experimenting with different audio/speech signals at different sampling frequencies, as well as different parameters for the AEC itself. This section describes those parameters and gives some guidelines on how to choose their values for optimum performance. The command line arguments correspond to internal variables with the same names. This makes real-time operation and integration into other applications as simple as directly assigning values to those variables in the source file and recompiling. The calling syntax is as following:

pbfdafaec.exe [/ARG <value>] [ /ARG <value>] [à]

All command line arguments have default values, so no need to list them all if you are satisfied with the default. The order of the arguments in the command line is not important. As an example, the following command

The success of the AEC in removing the echo from the Mic signal depends on many (inter-dependent) factors. To start with, for the adaptive filter to make a good estimation of the echo, the filter should have sufficient number of coefficients. The number of coefficients has direct relation with the amount of echo reduction (dB reduction). Correctly choosing the number of filter coefficients is, therefore, very important. Shorter filters result in low echo reduction and longer filters result in slow convergence.

6.3 stepSize

The second parameter affecting the AEC performance is the convergence time. This is the time needed for the filter to adjust itself to reach a reasonable approximation of the room impulse response between Spkr and Mic, and consequently, resulting in a reasonable echo reduction. The convergence time depends on factors such as the statistics of the FES input signal, the number of filter coefficients, and the algorithm used to adapt the filter.

In systems where blocks of samples are read and written to the outside world, the algorithm delay is integrated in the input/output process. This is the case for instance when a decoder is used before the AEC, which works on blocks of, samples or when a non real-time operating system such as Windows (or Linux) is used. In the latter case, the audio driver captures a buffer of samples and returns it to the application when the buffer is full. Playing the sound is also done whenever the application is ready with filling a buffer. In those cases, the algorithm delay is buried into the buffering process.

6.7 print

This parameter is a flag specifying whether to print the filter coefficients to a text file or not. A value of 0 will cause PBFDAFAEC not to save the coefficients. A value of 1 will cause the coefficients to be saved into a file called aeccoef.txt. The file will be overwritten if already existed.

The performance of the PBFDAF AEC is shown in Fig.2. This result has been obtained by running PBFDAFAEC with FES = aecfes.wav, NES = aecnes.wav, and RES=aecres.wav, filterLength = 64, blockLen = 64, partitions = 8, and stepSize = 0.1. You will find the above three wave files in the wav directory where you installed the software package. The top graph in Fig.2 is the estimated and actual impulse responses between the loudspeaker and the microphone. The middle graph in Fig. 2 shows two speech signals, the NES, (echo or microphone signal) in blue and the echo canceller output RES, (residual) in green. The bottom graph is the energy of RES signals divided by the energy of NES and the result expressed in dB, which indicates how fast the filter converges. Note however that this quantity is meaningful only when the NES is non-zero, in other words, when there is speech to cancel.

Commercial acoustic echo cancellers provided by DSP ALGORITHMS are much more complex than the one implemented in this simple version. A commercial AEC includes the following components besides the adaptive filter as shown in Fig. 3;

• Double Talk Detector (DTD): To detect when the two conversation parties are talking at the same time and stop the adaptation of the adaptive filter to avoid divergence.
• Far End Speech Detector (FESD): To detect when the remote speaker is active.. The adaptive filter is allowed to adapt only when the remote speaker is active and the local speaker is not active. In this case there is also echo to be canceled.
• Near End Speech Detector (NESD): To detect when the local speaker is active. When only the local speaker is active, the adaptation of the filter must be prohibited.
• Control Unit (CU): The outputs of the DTD, FESD, and NESD are analyzed in the control unit which generates a command to the adaptive filter defining whether the filter should calculate the estimated echo, update the filter, do both, or neither.
• Non-Linear Processor (NLP): When neither speakers is active, there is no need to do any calculations and the control unit generates a command to activate the NLP located in the residual path AGC unit to output a suitable signal.
• Automatic Gain Control (AGC): Two AGC units are used to regulate the sound level sent to the speaker (not shown in Fig. 2) and that sent to the communication channel. This is important to avoid communication quality loss when the speakers are far from the microphones or when they are moving while talking.
• Noise Canceller Unit (NCU): This unit is optional and can be integrated in the acoustic echo canceller or purchased separately. Its function is to clean the microphone signal from back ground noise and noise due to the use of cheap microphone and/or hardware.

If you are interested in test driving the full version of the acoustic echo canceller, or you have any questions or comments on this document or the software described herein, please email to support@dspalgorithms.com. We will be glad to help you.