By Paul Wheeler, Analog Devices Inc.

The technology landscape in automotive entertainment systems is becoming increasingly complex. Simple audio systems that distributed audio data across copper cable are becoming a thing of the past. To meet the requirements of multi-channel audio processing and distributed video, sophisticated network processing is becoming more prevalent. In particular, the MOST (Media Oriented System Transport) optical network with its associated DTCP encryption/decryption methodology is becoming adopted in many high- and mid-end vehicles. This trend, combined with the fact that automotive audio systems typically have to accommodate a wider variety of input sources (AM/FM tuner, CD, DVD drive, cellular phone, navigation system input) at varying sample frequencies applies increased pressure to DSP suppliers to provide higher performance and more highly integrated processors.

Typical MOST-Based Automotive High-End Entertainment System

The MOST bus was developed specifically to address the needs of the rigorous automotive environment. Based on optical fiber, this new network can support 24.8 Mbps of data, with the added advantages of less weight and reduced EMI characteristics than it’s copper predecessor.

Figure 1: Typical Automotive Entertainment system based on MOST

The MOST bus is based on a ring topology, which allows sharing of data across multiple sources and sinks. Data acquisition is facilitated by the MOST bus master (normally located in the head unit), and the network can support multi-master topologies, with a maximum of up to 64 devices on one network. To ensure security of data, the bus master will interrogate each slave on the bus and perform an Automatic Key Exchange (AKE) at power up. If the slave has a valid key for the bus, it is then allowed to source and sink data on the MOST bus using a predetermined protocol.

The MOST transfer protocol is made up of blocks that are split into frames. Each frame contains streaming data, packet data and control data. Streaming data is synchronized to the MOST clock and is continuously circling the network. Packet data is asynchronous to the MOST clock, and is generated on a need basis, an example of which would be an e-mail from a wireless PDA device. The amount of bandwidth allocated between the streaming data and the packet data is variable within the frame to meet the needs of the system at that specific time, and the control word contains stream information such as data type, where to find the data in the frame, size, and so forth. The control information may be distributed across multiple frames, and should be regenerated at the sink device.

Audio Processing in a MOST-Based Automotive Entertainment System

Figure 2 illustrates a simple MOST-based automotive audio entertainment system. The audio source, such as PCM, AC3 and DTS contents are taken from the DVD player via an SPDIF link to the head unit. The SPDIF link will operate at the sample frequency of the source (FS_in1), for example 44.1 kHz for CD-Audio or 48 kHz for DVD Video contents such as AC3 and DTS. As this encoded audio data is going to be passed over a network, the contents much be encrypted before transmission to prevent pirate copying of the original content. The encryption mechanism of choice for automotive systems is currently Data Transmission Content Protection (DTCP), which will be described below. The Analog Devices BlackFin architecture, for example, is well suited to this function with it’s rich peripheral set and optimized instruction set, which enable it to perform microcontroller like tasks along with traditional digital signal processing tasks. At the same time, the navigation system announcements must also be transported across the MOST bus to the amplifier to allow the driver to hear commands while driving. These PCM based signals are normally based on a 12.24 kHz signal, which we will call FS_in2.

The MOST transceivers take these various audio sources and rearrange the data into blocks to transfer on the bus (as described above). The audio packets, some of which may be DTCP encrypted (as FS_in1) are passed along the bus to the amplifier section, which normally resides in the rear of the car (see Figure 3). As the audio source data has been sent across MOST, the DSP must reconstruct the original packet data, and in the case of DTCP encrypted data, decrypt the stream to its original form. A side effect of transmission across the MOST bus is that the original sample rate of the source audio has been lost. Even with clock reconstruction techniques, the original source sample rate cannot be accurately reconstructed, which can result in audible pops and loss of sound.

To further increase the complexity of the system, encryption techniques using DTCP have become a requirement in networked applications to provide security to the digital data going across the network. DTCP has four layers of copy protection:

- CCI : Copy Control Information

- AKE : Device Authentication and Key Exchange

- Content Encryption

- System renewability

The CCI is based on the content to be transferred across the network and is determined by the content owners, for example “copy free,” “copy never,” “copy no more” and “copy once.” Before any content can be exchanged, the devices on the network must determine if they are authentic. There are two levels of authentification: Full Authentification and Restricted Access Authentification. After keys have been exchanged, content can be transferred across the network. The content is encrypted/decrypted using a predetermined base cipher engine and placed into a Protected Content Packet in the MOST transfer protocol. This protected packet has a header signature to identify that the content has been encrypted.

Solutions to Next-Generation System Issues

How are companies addressing the complex system issues of automotive network based entertainment systems? In the case of Analog Devices, the company has developed the SHARC ADSP-21365 processor.

Figure 4: Analog Devices ADSP-21365 SHARC processor for automotive entertainment

The ADSP-21365 is a 32-/40-bit floating point SIMD (Single Instruction Multiple Data) signal processor. With 4-Mb of ROM onboard, all of the mulitchannel decoder standards such as Dolby Digital, DTS and post processing modules including DPL2x, Neo6, and so forth are all supported. Customer specific post processing can be executed in the 3-Mb internal RAM space and, with audio specific development tools such as Visual Audio from Analog Devices, customers can increase their post processing portfolio with short design times.

To address the issues of multiple audio sources running at different base sample rates, Analog Devices has integrated the AD1896 discrete sample rate converter into the ADSP-21365. With eight channels of sample rate conversion and up to 140 dB of performance, multiple audio sources can be merged with zero memory and MIPS overhead, and all output post processing can be run at a single sample rate to further reduce the complexity of the data flow.

Other audio specific peripherals include six serial ports with native support of TDM and I2S, and integrated SPDIF Tx/Rx for direct connection to digital audio sources. The ADSP-21365 Sharc DSP also includes a hardware based DTCP M6 cipher engine that is DTLA compliant. The peripheral has two dedicated DMA buses to allow high speed transfer to and from the M6 engine without intervention from the core and, with native support of encrypt and decrypt, the ADSP-21365 enables a simple design path to full DTCP compliant systems. The engine includes functionality to enable dynamic update of keys. Using the on-board timers, the user can set a time period in which keys can be updated and changed to increase security across the network.

Audio processing involves intensive use of FIR and IIR filters. In recursive calculations, the quantization error due to the digital representation of the signals can cause a degradation of the audio quality. High-end audio processors, such as the SHARC processors, use a floating point representation for audio signals to reduce this error.

In high-end audio systems, the quality of the sound is generally measured by how accurately the low amplitude, or very quiet sounds, can be reproduced. As the amplitude of an audio signal gets smaller, the ability of a fixed-point processor to accurately reproduce this signal is limited, but for floating-point processors the accuracy with which the audio level can be maintained is contained within a fixed boundary, with a minimum SNR of 186 dB. With native support of 40-bit floating point precision, and 80-bit accumulators, the SHARC processor can deliver excellent audio performance.

Figure 5: SNR values for fixed and floating point processors.

Another important feature for a high-end audio processor is dynamic range. Dynamic range is defined as the ratio of the minimum to the maximum signal amplitude that the audio processor can reproduce without underflow or overflow. Once again, floating-point processors far exceed the limits of the fixed-point processor.

Figure 6: Dynamic Range comparison for floating-point and fixed-point processors.

With the increasing complexity of pre-decoder algorithms and post decoder algorithms, the number of MIPS, or execution cycles needed to complete the many combinations needed for a high-end audio experience is forever on the increase.

To combat these issues, the obvious answer is to increase the clock frequency of the signal processor. Due to silicon process constraints, there are many obstacles to this method, which has lead signal processor vendors to approach the problem with architectural improvements. Some signal processor vendors have used a MIMD architecture approach, which involves executing multiple instructions in a single cycle while performing multiple data moves. The architecture requires more memory, which directly influences the chip cost. The SHARC processor architects took the novel approach of a SIMD architecture, in which the same instruction can be used to implicitly exercise a second parallel arithmetic unit. The result is that the code size is dense so there is a reduction in the MIPS requirement to perform the algorithms. Due to this SIMD architecture, the audio signal processor can operate on stereo signals in parallel without any extra processing overhead. The SHARC core is based on a 5-stage code pipeline that is fully interlocked, which means that programmers can write code without worrying about when data will be available. The arithmetic pipeline is optimized to be one cycle, which means that the results of a calculation are immeadiately available in the next cycle for further computation. With automotive audio specific peripherals and a SIMD-based, 32-bit floating point core, the ADSP-21365 Sharc processor enables new levels of audio system integration

Custom Audio Post Processing Design using Visual Audio

The historical challenge faced by DSP users has been the development of software that makes optimum use of processor clock cycles and efficient use of memory. The long-used and laborious approach of hand-coding audio signal processing algorithms in assembly language has become less and less viable. This is particularly true when a large portion of the required effort goes into creating standard “checklist” and “me-too” features instead of focusing on differentiating the product with value-added features. A better approach to developing audio product software has been required. To fulfill this need, ADI has developed a graphical environment—VisualAudio—as an aid to designing and developing audio systems that use the SHARC processor family. VisualAudio provides audio system developers with most of the software building blocks—together with an intuitive, graphical interface, shown in Figure 7—for designing, developing, tuning, and testing audio systems.

Figure 7: Example of VisualAudio graphical-interface screens.

VisualAudio comprises a PC-based graphical user interface (GUI, the graphical tool), a DSP kernel (the framework), and an extensible library of audio algorithms (audio modules). Working in conjunction with ADI’s VisualDSP++ integrated development and debugging environment (IDDE) , VisualAudio generates product-ready code that is optimized for both millions of instructions per second (MIPS) and memory usage. By simplifying the process of developing complex digital signal-processing software, VisualAudio reduces development cost, risk, and time. As a result, audio-system developers are able to focus on value added by differentiating their audio products from the competition.

The Visual Audio tool allows system designers to focus on the development of custom post processing modules using an intuitive graphical tool, which, in partnership with the powerful SHARC architecture and the on-chip ROM decoder functions, allows fast and trouble free system development and product deployment.

Summary

The requirements of next-generation audio systems are constantly increasing, requiring both higher speed and more highly integrated audio processors. In order to maintain a viable market position, audio processor providers have to offer high performance components that enable their customers to keep ahead of the “audio feature curve,” while at the same time offering an easy-to-use suite of development tools that decreases customers’ time to market. Analog Devices approach to this challenge has been to introduce the third generation of 32/40-bit floating-point SHARC audio processors that provide high performance and memory/peripheral integration. Additionally, the company’s Visual Audio development tools simplify audio algorithm development by generating production-ready code for a large variety of common audio-processing modules.

References:

1. IEC 61937: Digital Audio - Interface for non-linear PCM encoded audio bitstreams applying IEC 60958

2. White Paper: An Overview of the Coherent Acoustics Coding System, Mike Smyth, June 1999

3. ADSP-21367/8/9 Data Sheet: Analog Devices home page, www.analog.com

4. White Paper: Designing Memory Efficient Real Time Audio Systems with Visual Audio, by Paul Beckman and Vincent Fung. (Analog Devices)

5. Advanced DSP Supports Home Theater and Advanced Audio: Paul Wheeler and Sharon St.Ours

Web Based References:

1. http://www.analog.com/dsp

2. http://www.analog.com/processors/communities/audioMusic/index.html

3. http://www.dtcp.com

Paul Wheeler is the SHARC DSP Systems and Applications Manager at Analog Devices

Acknowledgements: Thanks go out to Jasmin Infotech in Chennai and the ARTC team in San Jose for their knowledge of audio and their drive to push the boundaries of audio processing. Special thanks also go to Paul Beckman and the SHARC application team for all their support.