In-vehicle signal processing and audio system using TMS320C54x hands-free development platform

1. Overview of Hands-Free Applications

Why use a hands-free cellular phone system? Let's take a look at the following conversation to clearly understand how inconvenient the use of existing cellular phones can be in certain situations: "Excuse me, dear, I have to make a sharp turn." "Wait, I have to make a few turns." This inconvenience is caused by the handheld terminal of the existing cellular phone system. Drivers often have to put down their phones and drive with both hands, such as turning, and then turn back to talk. Interruptions in phone conversations are inconvenient and can even cause economic losses because mobile telecommunications charges are quite high. Another important issue is safety. Imagine a driver who uses only one hand to operate the steering wheel and is talking on the phone. Since the driver cannot operate the steering wheel with both hands, it is meaningless to have an anti-lock brake system and airbags. Therefore, hands-free cellular phone systems are becoming a must-have for drivers using mobile phones.

2. HFK Development Platform

The HFK Development Platform is a set of solutions including DSP, providing software and hardware design, enabling rapid development of the final product, and making it unique. The HFK development platform can be connected to TI's software development environment Code Composer StudioTM (CCStudio) development tool via JTAG. This development environment combined with documentation allows for rapid integration of TI DSP third-party software and accelerates product launch. The HFK development platform

is ideal for high-quality aftermarket hands-free kits that require high cost and performance. The HFK development platform is an ideal solution for HFK with Bluetooth capabilities.

Figure 1: Hands-free kit development platform structure diagram

HFK mainly includes six modules, namely:

- Digital signal processor;
- Audio codec;
- RF (FM) transmitter;
- Programmable logic device (PLD);
- Bluetooth transceiver;
- On-board power supply.

For detailed information and related documents, please refer to the following link: http://focus.ti.com/docs/toolsw/folders/print/tmdshfk5407.html

3. Echo cancellation software for hands-free system

One of the disadvantages of in-car hands-free radio/telephone systems is that the far-end speaker will feel an echo. In order to achieve comfortable full-duplex hands-free calls in the car, one of the most important software elements is the acoustic echo canceller (AEC). The European Telecommunications Standards Institute (ETSI) is currently developing standards for AEC systems.

The echo phenomenon is caused by the coupling between the loudspeaker and the loudspeaker. In full-duplex communication, there will be a delay when the far-end speaker hears its own voice. The length of the delay depends on the delay between the car and the Global System for Mobile Communications (GSM). The echo path length is a critical parameter for AEC.

Adaptive filtering (more precisely the NLMS algorithm) is one of the most common solutions for AEC. The NLMS algorithm provides a good compromise between computational load and performance. [page]

Another problem with AEC is the fuzzy tone (DT) situation where two people are talking at the same time. DT can cause the adaptive algorithm to diverge if not detected.

The AEC software uses the NLMS algorithm to cancel the echo, executing the C54x DSP assembler.

NLMS Algorithm

The NLMS algorithm updates the coefficients of an adaptive finite impulse response (FIR) filter, which is used to predict the echo. The prediction is then subtracted from the actual echo, giving the residual echo.

Active Channel Detection

A key feature of the AEC algorithm is active channel detection. When the far-end operator is silent and the near-end operator is talking, the filter cannot be adapted because the near-end operator is no longer the echo. Active channel detection is achieved by calculating the signal energy and comparing it to an adaptive threshold.

Dull Tone (DT) Detection

In the DT case, the near-end signal at the loudspeaker consists of echo and near-end speech (i.e., dummy tone). The residual error used to update the filter coefficients includes the near-end speech, and if the algorithm is still adapting, the algorithm may start to diverge, which must be avoided. DT detection uses an energy-based algorithm with a variable threshold to address this problem. The

baseline

AEC software benchmark (expressed in 16-bit words) is:

- Code size: 154 words;
- Static RAM: 527 words;
- Write RAM: 2 words;
- Maximum computational cost is 4.7MIPS.

The computational cost is highest during the ST period; it drops to 2.4MIPS during the DT period. The ST period represents the majority of the conversation, while the DT period only occurs in shorter, limited cases.

4. CVC-HFK Software

Figure 2: CVC-HFK application diagram

CVC-HFK (Clear Voice Capture - Hands-Free Kit) is an optimized HFK solution that integrates echo cancellation, noise suppression, non-linear processing and other functions. The CVC-HFK solution uses a comprehensive adaptive sub-band approach to improve key aspects of performance while keeping resource costs low. In an automotive environment, ambient noise is a major issue for hands-free systems. Therefore, in addition to echo cancellation, Clarity CVC-HFK also provides an integrated One Microphone Solution (OMS) noise suppression algorithm. The OMS solution supports adaptive noise cancellation, which reduces ambient noise in the microphone signal (incoming), extracts the desired speech, and transmits the clean speech (outgoing) to the far-end user. Since CVC-HFK is fully adaptive, there is no need for excessive adjustments. Below, we will introduce the CVC-HFK solution and its key aspects of performance.

CVC-HFK AEC

The CVC-HFK echo canceller is a "stateless" AEC that uses a variation of the standard frequency domain NLMS algorithm as its main adaptive filter. We will explain the advantages of using these methods below. First, the sub-band frequency domain approach de-correlates or whitens the input signal in each band, which allows for faster convergence than a comparable time domain AEC. Second, the stateless AEC allows for continuous filter adaptation, which improves robustness in noisy environments and overall fuzzy sound performance. We recall that the loudspeaker signal in the DT case contains both echo and near-end speech. Near-end speech is not correlated with the echo signal, which can cause adaptive filter divergence if there is no process to avoid it. Third, NLMS allows for consistent convergence independent of input amplitude. As a result, the CVC-HFK AEC achieves a typical 40 dB ERLE (Echo Return Loss), a maximum of 50 dB ERLE, and a fast convergence time of around 80 ms, with full-duplex operation in most environments. In addition, the CVC-HFK AEC uses a 64 ms tail length for its adaptive filter, which allows for greater flexibility in terms of internal capacity.

CVC-HFK NS

The CVC-HFK Noise Suppressor is a frequency domain algorithm that uses the characteristics of speech and noise to help extract speech from synthetic noise and speech signals. The two main modules of CVC-HFK NS are speech component analysis and speech extraction.

The speech component analysis module uses the temporal and correlation properties of speech and noise to build a predictable model of speech components. The speech extraction block modifies the frequency components based on the speech and noise model. In addition, the speech extraction block takes advantage of the principles of acoustic quality to minimize the noise floor and perceived speech distortion.

The CVC-HFK NS uses this approach to achieve a 10-15dB SNR (signal-to-noise ratio) improvement in noisy environments while maintaining good speech quality. In very low noise environments where the SNR is already high enough, no speech distortion occurs because the NS is turned off.

CVC-HFK NLP

is minimal due to the increase in system distortion. The amount of distortion added by the CVC-HFK NLP is much lower than that of standard NLP modules such as the center clipper because it uses information from the input and error signals to determine the additional attenuation.

Since all CVC-HFK modules use frequency domain algorithms, they can significantly save memory and simplify computational complexity compared to solutions that use both time and frequency domain algorithms. [page]

5. System Integration Design

When integrating the TI-HFK board with a cellular hands-free kit, several components and appropriate interfaces are required to achieve good mobile calls.

You must select components that are compatible with both the CVC-HFK application software and the board hardware to achieve good performance. The HFK supports a variety of different microphones, speakers, and car audio systems. However, to reduce changes to the application manual, we have selected dedicated industry-standard components that will greatly help you make successful adjustments. Three connections are required from the TI-HFK board to the cellular kit to achieve integration:

the microphone input to the TI HFK board;
the outgoing, processed audio output; and
the input, the signal received from the cellular kit.
Here are some recommendations for speaker and loudspeaker placement design.

Loudspeaker Position and Orientation

To achieve the best overall loudspeaker performance, there are some key variables that should be understood before the final installation of the device in the vehicle.

It is recommended to maintain a distance of 46 cm (18 inches) between the loudspeaker and the user's mouth in the vehicle. The recommended distance ranges from 30 to 56 cm (12 - 22 inches).
Avoid exposing the loudspeaker to drafts (windows and fans) as much as possible;
consider the size of the loudspeaker and the mounting scheme so that the front of the loudspeaker can be aimed at the user's mouth in the vehicle.

Based on these considerations, you can choose the best loudspeaker location with the help of Figure 3. First, follow the recommendations in priority areas 1, 2, and 3. Once you have made your decision, you can secure the loudspeaker with a metal plate or Velcro strap, and you can terminate the cable back to the electronic device. Then keep the cable hidden and maintain the appearance, and keep the cable fixed and not pinched or tangled. Finally, avoid parallel cables connected to antenna connectors and other noisy cables.

Figure 3: Loudspeaker placement

Loudspeaker Positioning

It is recommended that the loudspeaker be positioned to provide good acoustic performance while not interfering with the loudspeaker pickup zone. The pickup zone is a +30 degree cone, extending outward from the front of the loudspeaker to the driver of the vehicle.

The loudspeaker should be located at least 1 meter (3 feet) from the loudspeaker. The loudspeaker should be positioned away from the loudspeaker pickup zone to reduce the chance of echo feedback. Ideally, the loudspeaker should be located behind or 90 degrees from the front of the loudspeaker.

The level of acoustic distortion of the loudspeaker will have a direct negative impact on the echo event.

6. Conclusion

The use of hands-free systems continues to grow in popularity, and users expect ever-increasing performance. Given the many options available for HFK implementation, it is clear that the integration of software algorithms and hardware signal processors is a thoughtful move that, when properly implemented, will be of great benefit. The HFK development kit that addresses all of the above issues will provide clear benefits to anyone interested in developing or marketing such products. The TI TMS320C5407 development kit with AEC and CVC-HFK has the performance and flexibility needed to quickly and inexpensively bring HFK solutions to market.

References
1. spru703 "Texas Instruments Car Hands-Free Kit Development Platform User Guide", September 2003;
2. sprt289a "Car Hands-Free Kit (HFK) Development Platform Product Brief", June 2004;
3. spra162 "Echo Cancellation Software for Car Hands-Free Wireless Systems", July 1997;