Home-theatre systems, AV, and DVD receivers have become much more complex since the introduction of digital audio. As analogue hi-fi components continue to be used, these audio hubs must seamlessly interoperate with both analogue and digital equipment. Today, the migration from stereo to multichannel audio further increases complexity. Modern AV receivers must support a variety of formats, such as 5.1, 6.1, or 7.1 channels, as well as traditional stereo, using analogue and digital signal sources. Fortunately, Moore's Law and advances in mixed-signal IC design have made this inherent complexity transparent to the end user while shrinking the physical size of such systems.
The first consideration facing designers is system partitioning. Early multichannel systems were often built by arranging several instances of an existing stereo solution in parallel. The drawback of this architecture is that it disregards the differences between the various channels. Firstly, the subwoofer channel has a reduced bandwidth. Its signal is often generated in real time by summing the low-frequency components of the full-bandwidth channels. Secondly, when a stereo source is played through a multichannel system, only the front left (FL) and front right (FR) signals are provided by the source; all remaining channels are generated from these two in a process known as multichannel expansion. The reverse process, known as down-mixing, is used when a multichannel signal is recorded onto a stereo medium. These mixing and filtering requirements are difficult to implement using traditional stereo components, where the individual channels are mutually independent.
One approach is to aim for a system-on-a-chip solution that integrates all functions on a single chip, requiring only passive external components. While this is perfectly feasible using today's mixed-signal CMOS technologies, it does not always yield the highest-performing or most cost-effective solution. For analogue circuitry, the ideal tradeoff between performance, die size, power dissipation, and cost is invariably obtained with larger process geometries than those best suited to digital circuit blocks. As a result, system architects are faced with a difficult choice between conflicting requirements. Using a more advanced fabrication process means that lower supply voltages must be used. This reduces the voltage swing available for analogue elements, worsening their signal-to-noise ratio (SNR). Moreover, the characterisation of new processes is often incomplete, making it difficult to run the accurate simulations necessary to optimise the performance of analogue circuits. Conversely, using an older process with larger transistors needlessly increases the size and power dissipation of digital circuit blocks.
An alternative proposition is to partition the system into a mixed-signal device and a digital signal processor (DSP). This approach makes it possible to optimise each component separately, so that demanding performance targets can be met more easily. Moreover, DSP vendors often provide an extensive code base of algorithms as well as programming tools, simplifying the deployment and customisation of signal-processing functions. This leaves mixed-signal vendors free to concentrate on their core competency and offer a "just add DSP" solution. Such ICs typically include a number of DACs and ADCs (digital-to-analogue / analogue-to-digital converters) as well as analogue mixing and interfacing functions.
DIGITAL CONTROL FOR ANALOGUE SIGNALS
Audio playback in a multichannel system requires volume and tone control, as well as bass management—the process of redirecting low frequencies to the subwoofer. Providing these functions for both analogue and digital source signals can be difficult in systems where audio performance is important. In a digital implementation, any analogue signal must first be digitised, then processed and finally converted to the analogue domain before being passed to an amplifier. The extra analogue-to-digital conversion adds cost, and inevitably introduces a certain amount of noise into the signal.
Implementing signal-processing functions in the analogue domain does not add any conversion steps, because digital signals are eventually converted to analogue in any case. However, once a signal is in the analogue domain, managing subsequent signal-processing functions becomes more difficult. Traditional potentiometers are not a viable option, because they cannot easily be controlled by a microprocessor or remote control. Replacing them with digital potentiometers introduces harmonic distortion and is relatively costly.
A more elegant solution is to use programmable-gain amplifiers (PGAs), whose gain can be digitally controlled. Low-noise PGAs are a classic example of mixed-signal circuitry that lends itself to integration on one IC along with DACs and ADCs. They also achieve very high performance while allowing for full digital control. By splitting the signal into different frequency bands using simple RC filters, the gain of each band can be set independently—a digitally managed tone control. For bass management, low-frequency signals are separated out of each channel with a simple RC low-pass, and summed together to form the subwoofer signal.
CLOCKING AND INTERFACES
Besides local audio sources, modern AV receivers should also accept signals from remote digital sources, such as a CD player connected via an optical cable. The S/PDIF standard, typically used for optical connections, transmits audio data and associated clocks in a single bi-phase encoded signal. At the receiving end, a clean clock must be recovered from the combined signal. This requires a low-noise PLL (phase-locked loop), another mixed-signal function that can be integrated. Since the clock recovered from the S/PDIF signal was originally generated by the remote system at the far end of the optical link, it is not synchronised with the local system clock. In order to avoid digitally re-sampling the audio data to the local clock frequency—an operation requiring considerable computing power—audio DACs should be able to run on the S/PDIF clock as well as the local clock.
A related problem occurs when two different clocks are used at the same time, as when playing a signal from one source and simultaneously recording to a different medium. For example, traditional audio CDs are designed to deliver data at a sampling rate of 44.1kHz, while most other digital media sample at 48kHz or at multiples of that rate. The gap between such incompatible clock domains can be overcome without digital re-sampling, by converting signals to the analogue domain and re-digitising them at a different sampling rate. Two conditions must be met to make this possible. Firstly, the audio IC must be able to internally run ADCs and DACs on two different clocks. Secondly, it must provide a way to input and output data at different rates. Although I2S and other similar audio interfaces can handle simultaneous playback and recording, they use a single clock for both data streams. As a result, two separate audio interfaces are needed to provide the necessary clocking flexibility.
Full-featured systems have up to three audio interfaces—a primary interface between the DSP chip and the mixed-signal IC, a secondary interface for local digital sources using a different clock, and an S/PDIF interface for remote digital sources. Depending on the usage scenario, the DACs and ADCs within the mixed-signal ICs take their clock signal and data from one of these three interfaces. Additionally, a number of analogue inputs and outputs are generally provided in order to support legacy analogue media.
DETERMINING AUDIO PERFORMANCE
Partitioning complex audio systems into purely digital and mixed-signal parts frees designers from a number of delicate tradeoffs, enabling them to bring the full benefits of both technologies to the end product. Pure digital DSP chips can thus be manufactured with a higher degree of integration, reducing their physical size, power consumption, and cost. Mixed-signal ICs, on the other hand, can use the larger transistor geometries and higher supply voltages necessary to deliver a truly low-noise, low-distortion audio signal.
This system architecture accommodates both analogue and digital signal sources without introducing extra signal conversion steps. Unnecessary losses of signal quality are thus avoided, while offering full digital control over every part of the signal chain. Despite the split into two main ICs, chip count can be reduced by integrating ancillary functions such as volume and tone controls, clock generators, PLLs, or a complete S/PDIF interface. Paradoxically, as single-chip solutions often require additional components for such functions, the actual chip count is typically lower in a two-chip system.