Audio ADCs hit the high notes: everything from filter performance and software to clock signals and PCB layouts affects audio-ADC performance.
"To start, we need to know about the application and its requirements," said James Scanlan, Americas applications manager at Wolfson Microelectronics, a supplier of high-end ADCs, DACs, and codecs. "Is it a voice product? Does it need high fidelity? Will it operate in a noisy environment? Must it conserve power? What are the signal-to-noise ratio [SNR] and the total harmonic distortion [THD] requirements? Answers to those and similar questions help us guide designers to specific ADCs and CODECs."
"Say you produce high-end automotive stereo systems," said Scanlan. "You have a head unit in the car's dashboard that puts out four channels of audio in analog form. The car-trunk amplifier's ADCs sample the audio signals, process them, and drive speakers. In that case, you need ADCs with a high SNR, but ADC power consumption becomes less important."
According to Bob Adams, a Fellow and manager of audio development at Analog Devices, ADCs designed for high-end audio applications now almost exclusively use the sigma-delta conversion technique. "And many engineers no longer use discrete ADCs, but design with audio codecs instead. You can get from one to eight ADCs and two to eight DACs in one codec package. The dynamic range of these codecs can be as high as 100 dB, which makes them a suitable choice for many consumer applications. Many of these devices also offer low power consumption for portable devices. If designs require higher performance and can tolerate somewhat higher power consumption, there are devices with dynamic-range specs that approach 120 dB, which make them suitable for the professional and automotive audio markets."
When Do 24 Bits Equal 16?
"Many engineers who have used an ADC for industrial measurements can get a bit confused when we talk about audio sigma-delta ADCs with 24-bit resolution," said Dafydd Roche, home and audio pro product marketing manager at Texas Instruments. "Often people think a 24-bit ADC means they will get 24 bits of useful information. But in some cases a 24-bit converter can perform worse than a regular 16-bit converter. A 24-bit ADC that has a 96-dB dynamic range produces about 16 bits of real data. Even though you get 24 bits in the ADC's serial output stream, you have only 15 or 16 bits of real data and the rest of the bits are noise or zeros."
"The noise inherent in CMOS devices limits ADC performance to less than 24 bits, or less than 144 dB," noted Robert Martin, a system engineer for audio converter products at TI. "1 don't believe any ADC today gets close to 144 dB. Engineers want to squeeze every decibel of performance out of an ADC, but they might inadvertently compromise their ADC's performance by including pre-amps, filters, or other analog stages in their front-end circuits. If they specify a high performance ADC and then use low-performance op-amps in those front-end circuits, that asks for trouble. The ADC input circuitry must be carefully considered in order to achieve the best converter performance. Using a noisy op-amp to amplify a small audio signal might cause the amplified noise of the op-amp to degrade the ADC's performance. So it's like driving a Ferrari with the parking brake on: It's a great car but you can't go at top speed."
Not every application demands a high dynamic range. "We see a large market for low-cost speech devices in products such as intercoms, elevators, toys, and speaker phones," said Steven Marsh, strategic marketing manager with Microchip's Digital Signal Controller Division. "The audio spectrum for those devices ranges from zero to 8 kHz, as opposed to standard telephone-quality audio between zero and 3.3 kHz. Because cost is a key design influence, a 12-to-14-bit ADC--often within a controller chip--will suffice."
"Engineers frequently think they must use a 16-bit ADC in audio circuits," noted Sunil Fernandes, an applications engineer in Microchip's Digital Signal Controller Division. "But as they experiment with a 12-bit ADC, they discover that's not entirely true. They get good audio quality with only 12 bits. In a toy or intercom system, the characteristics of the speaker and microphone, rather than ADC resolution, often determine audio quality. So, a 12-bit ADC gives you a lot of 'head room' in these consumer-type products."
"People can take the MPLab Starter Kit for dsPIC DSCs and demonstrate how well a 12-bit ADC digitizes an audio signal," said Fernandes. "They can press a button, speak into a microphone, record 12-bit speech, and then play it back. The kit produces a PWM Class-D audio output they can listen to."
Oversampling Boosts Resolution
But suppose engineers decide they still need more resolution. "They can oversample an audio signal at 128 k samples/sec to improve resolution of a 12-bit ADC by 1.5 to 2 bits," explained Fernandes. "So your ADC can have an effective resolution of 13.5 or 14 bits. That's just one of the things designers can do within a dsPIC chip. Microchip has a variety of post-processing software libraries for functions such as equalization, filtering, spectral analysis, automatic gain control, noise-and-echo cancellation, and speech recognition."
In contrast to the sigma-delta ADCs used for many high-end audio applications, the dsPIC chips rely on a successive-approximation-register (SAR) ADC. Software can trigger SAR ADC conversions whereas a sigma-delta converter runs continuously. Although an SAR ADC usually requires a sample-and-hold circuit on its input, the Microchip dsPIC devices do not. The sample-and-hold comes built in.
A dsPIC-based circuit might need an anti-alias filter, though, depending on the application and sample rate. "Anti aliasing gets into the realm of digital filtering as well," said Marsh of Microchip. "Often engineers can replace a sophisticated analog filter with a simpler anti-aliasing filter and use digital-filter software. That approach significantly cuts the component count while it improves filter performance. A dsPIC-based circuit, for example, could combine the digital filter code with other wave-shaping software for an added benefit."
Look Beyond Resolution
Although ADC resolution first comes to mind when engineers look at ADCs, other performance characteristics deserve attention, too.
"Sigma-delta ADCs include a digital decimation filter that typically offers a linear-phase response through the audio pass band," explained TI's Martin. "Human ears will quickly detect out-of-phase audio signals. ADCs used with sensors in industrial applications usually don't require this type of phase linearity. So at first, engineers with industrial backgrounds might not appreciate the need for phase linearity in audio circuits."
Phase relationships become important at the signal-input side of ADCs, too. "We talk on our mobile phones in noisy environments," said Wolfson's Scanlan. "Typically a phone uses one microphone, but multiple microphones give engineers the opportunity to apply audio beam-steering techniques that help attenuate noise. But beam steering demands in-phase sound signals from those microphones. In this type of design you might have a discrete ADC next to each microphone to send digital data to a processor. By placing an ADC at each microphone, you no longer must run sensitive analog signals through digital and RF circuits that could add noise."
All ADC circuits must put up with some noise, and oversampling and noise shaping help decrease quantization noise. That type of noise arises normally in an ADC due to the quantization of a continuous signal into discrete samples. "Oversampling distributes quantization noise over a wider bandwidth--between DC and half the sampling frequency, while noise shaping moves a large chunk of the noise out of the Nyquist band," said TI's Martin.
Reduce Audio Noise
Acoustic noise also affects the acquisition of useful audio information. "But dealing with audio noise often involves more art than anything else," said Marsh. "Engineers have many ways to attenuate noise, but the methods they choose depend on the types of noise present and what the engineers need to achieve."
"You might think of noise as a high-frequency signal you can simply filter out," continued Marsh. "Noise suppression involves more than that. Suppose you must suppress noise from a factory area superimposed on someone's speech. You must first detect the speech and next determine when the speech isn't present. Then you sample the factory noise alone and use its characteristics to filter the acoustic noise out of the speech signal. That technique goes beyond a simple filter task." Microchip has noise suppression libraries that work with data from the 12-bit SAR ADCs in its dsPIC devices. These libraries now work with traditional telephone-quality signals, but the company expects to soon offer software that also operates on wide-band speech signals.
View this entire article online!
This article continues online with a look at spurious noise, clock circuits and jitter. Go to www.ecmag.com/cs-Audio-ADCs-Hitaspx.
by Jon Titus, Senior Technical Editor
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||COVER STORY|
|Publication:||ECN-Electronic Component News|
|Date:||Apr 1, 2009|
|Previous Article:||GaN-based power devices make possible a new paradigm in power conversion.|
|Next Article:||Standards for embedded microcontrollers.|