Abstract
It is known that audio system components, such as loudspeakers, amplifiers, and microphones, exhibit nonlinearities when exposed to high dynamic range audio signals. Thesecomponents can saturate, leading to irreversible signal loss. Since most echo cancellation
systems (ECS) operate in the digital domain, it is necessary to model and compensate
for these nonlinearities. However, this is challenging and can degrade the performance
of the ECS. Furthermore, nonlinear ECS often involve significant computational complexity, making them unsuitable for real-time applications.
To address these issues, an analog ECS can be employed. An analog ECS uses
the analog transmit signal as a reference and subtracts the replicated signal from the
received signal before it reaches the input amplifier of the receiver. This approach inherently models the nonlinearities of the transmitter within the reference signal, allowing
the use of linear adaptive filters for the audio channel. Additionally, this method prevents the saturation of the reception path by pre-cancelling the echo signal.
This thesis evaluates the performance of analog adaptive filters with quantized components, such as finite precision attenuators, for acoustic echo cancellation. The main
motivation behind using adaptive filters with fixed characteristics is that they can be
easily and cheaply built and are linear over a wide frequency range. It investigates
the filter’s performance under varying degrees of quantization and validates the results
through an audio prototype.
The proposed ECS employs an analog transversal filter and an analog Least-MeanSquare (LMS) algorithm for weight adaptation. The analog LMS algorithm, chosen for
its straightforward implementation and robustness to weight quantization, is enhanced
by dithering to linearize the update equation.
This thesis demonstrates that the analog ECS can outperform a digital ECS, particularly in scenarios involving highly autocorrelated and nonlinearly distorted desired
signals. The analog ECS achieves an average suppression of 16 dB with only 15 taps
and converges within 1.5 ms.
| Date of Award | 2024 |
|---|---|
| Original language | English (American) |
| Supervisor | Hans-Georg Brachtendorf (Supervisor) |