Dynamically improve echo and noise cancellation performance

Voice quality in mobile handsets, wireless networks, hands-free devices and other mobile communication systems is a key factor in establishing consumer preference. Echo and noise are inherent problems in wireless communications. Signal processing technology is needed to solve voice quality issues and ensure high-quality audio output that is acceptable to the market. The traditional approach is to use independent echo and noise cancellation modules at the near end or in the transmission path. This method performs well when the surrounding conditions remain unchanged, but if the surrounding environment changes, such as opening a door or loud noise, the audio system will have difficulty adapting to the changes and the audio performance will degrade.

The new approach combines echo cancellation, noise suppression and other sound enhancement technologies to dynamically adjust system parameters faster based on environmental changes. In most cases, we can complete the adjustment before consumers even notice any problems with the sound quality. Similarly, this new approach achieves higher levels of integration and can solve larger noise and echo problems, enabling full-duplex voice calls that sound very natural.

The dramatic advances in echo and noise cancellation technology are timely, as many U.S. states have enacted laws that completely or partially prohibit drivers from talking on mobile phones while driving. Similar laws are also in place in most European countries and many other countries around the world. The emergence of these regulations has further increased the demand for hands-free technology and requires effective noise and echo cancellation in the interior environment of the car, which is also the biggest design challenge of hands-free systems. Designers need easy-to-use hardware and software to provide hands-free audio products with the same sound quality as traditional handheld products in order to meet user needs.

Sources of echo in wireless communications

There are two main sources of echo in wireless systems: electrical echo and acoustic echo. Electrical echo occurs when poor design results in the speaker signal being coupled directly to the loudspeaker signal. The best solution to this problem is good design.

An even more challenging problem for us is acoustic echo. Acoustic echo occurs when an amplified loudspeaker signal echoes through a loudspeaker. Eliminating this echo is quite difficult and we have to take several factors into account. The amplified loudspeaker sound is reflected on multiple paths at different times. This indirect echo lags significantly behind the original signal, because sound travels through air at only 300 m/s, and the echo reflections are also distorted due to the added complexity of mechanical vibrations.

Half-duplex switching technology

Figure 1: Half-duplex solution

The most basic solution to the echo problem is to disable the near-end voice channel when far-end voice is detected. This eliminates acoustic echoes but only allows one person to speak at a time. For example, with traditional walkie-talkies, when you press the talk button, you can no longer hear what other people on the line are saying, so the rules of two-way radios require that the speaker must clearly indicate "done" when he or she has finished speaking. Later, new technology replaced the talk button with a voice activity detector (VAD), which automatically turns the near-end voice channel on and off when far-end voice is detected. In the early days of mobile communications, we were able to accept this limited technology, but as users gradually became accustomed to full-duplex wired calls, they will no longer accept this restrictive one-way calling technology in the future. Because full-duplex wired calling technology allows them to communicate freely, express ideas, agree or disagree with each other's views, and pause at any time without worrying about the loudspeaker suddenly becoming unusable.

Figure 2: Traditional echo cancellation technology

Here's why nearly all cell phones, hands-free devices, and phones with speakerphones offer some sort of echo cancellation technology. Today, nearly all devices cancel echo using the basic approach of monitoring the far-end signal and then subtracting it from the received signal. This would be easy to do if the amount of echo was known and constant. But in reality, the amplitude and duration of the echo depend on the environment in which the wireless device is used, and this environment often changes, so traditional echo cancellation technology requires continuous monitoring of the near-end and far-end signals. The acoustic echo canceller algorithm uses a reference signal from the near-end speaker to estimate the echo path and remove the echo from the near-end speaker signal.

The design and tuning of the adaptive filter is a decisive factor in the performance of echo cancellation. The filter usually uses known characteristics of the audio signal to calculate an estimate of the echo and adjusts the filter parameters accordingly to minimize the error. Echo cancellation is usually achieved by updating the filter coefficients using the Normalized Least Mean Square (NLMS) algorithm. This algorithm minimizes the mean square error of the canceller, which is the residual echo. Adaptation is usually normalized to the signal power to be independent of the signal level.

We can perform these calculations with sufficient accuracy in most cases to reduce the perceived echo. The problem is that the algorithm depends on the stability of the echo path between the speaker and the amplifier. The echo path changes whenever there is an object near the phone that blocks the sound (such as placing the phone on a table, touching the keypad, or covering the speaker with paper) or when the distance from the amplifier to the speaker changes (putting the wired microphone back in place). When the path changes, the algorithm has to adjust to the new echo path, which causes a delay. During the adaptive delay process, the echo is transmitted on the near-end signal path.

When designing an echo canceller, it is important to understand the environment in which the device will operate. Are the microphone and speaker in fixed positions? Will changes in position affect normal operation? What is the longest echo path allowed in the device's operating environment? How much noise is expected? Will the noise change (for example, in an automotive environment)? How loud should the device volume be? What is the echo return loss between the speaker and the microphone? How loud should the near-end talker's voice be compared to the echo at the microphone? Only by answering these questions can we design the best traditional echo canceller that can adjust to the known environment. However, when the environment changes, the filter coefficients are still adapting to the new echo path, and we already hear the echo. Depending on the initial parameter settings, the adaptation process takes 5 to 10 seconds.

In addition to the problem of echo affecting the near-end signal quality, background noise can also cause adverse effects. The solution to this problem is to use a noise canceller. Typical noise cancellers work independently of the echo canceller, and any interference issues can be ignored. Unlike echo cancellers, noise cancellers do not have a reference signal to rely on. It must either estimate the noise and cancel it from the speaker signal, or it can only estimate the speech. In either case, the noise should be targeted to maximize performance. Combining the control signal of the noise canceller with the AEC can achieve a more accurate voice activity detection environment and improve the integration effect. Without this interaction, the system may mistakenly cancel the speech signal as noise.

Figure 3: New approach integrates echo and noise cancellation with other audio processing techniques

To address the limitations of traditional technologies, we have developed a new approach to improve wireless audio quality (see Figure 3). The basic difference between the old and new approaches is that the new approach integrates echo cancellation, noise cancellation, and other audio signal processing functions, all controlled by a new full-duplex control module. This approach uses the same core NLMS algorithm, but has some specialized features that not only take advantage of the system technology breadth of this integrated approach, but also dynamically adjust system parameters to accelerate NLMS reintegration.

Full-duplex control is the key to the new approach's improved performance. By combining the audio portion of wireless communications with the latest digital signal processing, nonlinear control algorithms can be used to adjust for sudden environmental changes, such as a door slamming in the background or a sudden gesture or movement of the user's hand holding the phone. Sound quality is further improved by optimizing different control algorithms simultaneously under the main controller. Finally, the use of a more powerful signal processing architecture allows us to add new features, such as filling the background with natural-sounding comfort noise to compensate for changes in the noise floor and avoid noise pumping.

Integrating system processing techniques for all key elements of the near-end and far-end audio paths to optimize signal quality at both ends of a call was very difficult with previous DSP generations. Recent DSPs offer the right balance of performance and advanced on-chip memory, with algorithms sophisticated enough and audio processing integrated enough to accommodate rapid optimization of different audio components, helping to achieve the best wireless voice quality.

How the new method works

The new approach uses the entire system to understand the current working environment and dynamically adjust system parameters to achieve optimal performance. Analysis and parameter adjustment are the tasks of integrated full-duplex control. Full-duplex control technology evaluates near-end and far-end signals, first determining whether the signal is currently working, and then evaluating the signal quality from different perspectives. Based on this information, the full-duplex control mechanism will perform comprehensive dynamic adjustments to each module to improve the quality of the near-end and far-end signals.

Full duplex control on the near-end signal path controls the parameters of the nonlinear processor, echo canceller, and noise canceller to reduce echo and noise. Full duplex control on the far-end signal path controls the dynamics processing mechanism to adjust the audio signal to increase the volume output while reducing speaker nonlinearity. Graphic equalizers and sound quality enhancement techniques are used on both signal paths. Graphic equalizers are used to adjust the transmitter (speaker and amplifier) ​​and can also be used to adjust the frequency characteristics of the audio signal. Sound quality enhancement techniques are used to adjust the sound quality for optimal voice intelligibility.

The characteristic of using this system technology is that the full-duplex control technology uses the environmental information learned by the system to achieve higher volume and lower echo, and can quickly adapt to the changing environment.

Designing a new audio processing system

The integration of new audio processing systems has increased significantly, which has brought us a series of design challenges. First, we should find a suitable DSP that can provide the required high performance for new designs while providing a suitable programming environment to support designs that are much more complex than traditional echo and noise cancellation technologies, thereby shortening the design cycle of mobile communication systems.

For example, the C5000 DSP platform from Texas Instruments (TI) achieves an optimized combination of processing performance and large-capacity on-chip memory, which helps reduce the workload of off-chip memory and reduces the burden on the processor. The choice of architecture is very important. The architecture should not only be optimized for audio processing, but also include a rich range of devices to achieve the perfect combination of leading power saving features, rich peripheral selection and small packaging. TI has an extensive network of third-party developers that can provide a variety of products to help OEMs and ODMs add features such as audio streaming of MP3 and WMA files, Bluetooth, voice recognition, and phone book downloads.

The development work adopts the model-based design method, which can model the complex sound behavior of multiple devices and generate test vectors for the device environment. We use Simulink from MathWorks to design and develop the relevant models. Designers can create independent algorithm modules in C code and integrate them into the simulation environment for testing.

Engineers can use the models provided by the software to simulate the performance of echo and noise cancellation systems by simply writing the corresponding scripts to describe typical working conditions. This approach allows us to evaluate multiple design solutions and further understand the performance while saving design time and cost. Designers can quickly modify the model and observe the performance changes, so as to quickly optimize the design solution and achieve the best audio performance.

Once engineers are satisfied with the system simulation results, they can generate C binary code for the TMS320C5000™ DSP platform. We use the Code Composer Studio™ integrated development environment to easily create binary object code, debug the source code while testing the binary image, and help engineers easily debug the design. The combination of modeling technology and Code Composer Studio target support allows engineers to verify the performance effectiveness of the design with simulation input on actual hardware. Later, they can also use real-time audio input and output independent of the simulation model to make further fine-tuning and further optimize the code during evaluation.

The ability to achieve optimal performance of echo cancellation and noise cancellation depends on the ability of the system solution to dynamically adapt to the changing environment. Dynamic adjustment of system parameters should respond quickly to changes in the environment to avoid intermittent echo and noise interfering with the current generation technology. Testing of the above solutions can only be done with a good modeling environment. The key to a successful end product is to choose the right DSP technology that not only provides powerful signal processing capabilities, but also provides a development infrastructure to ensure timely market launch within a certain period of time.