John Siau received a BSEE from Syracuse University in 1980, and has 32 years of engineering design experience in audio, video and telecommunications systems. He holds two patents for video image stabilization systems. He has designed numerous A/D and D/A converters, low-jitter phase locked loops, digital signal processors, and FPGA cores. John is Vice President and an owner of Benchmark Media Systems. He designed all of Benchmark’s converters and their two new microphone preamplifiers.
Benchmark has been providing leading-edge performance to broadcasters, recording studios, sound reinforcement contractors, and hi-fi enthusiasts since 1983 and their products have won numerous awards in both the professional and home audio industries. For over 25 years, all Benchmark products have been manufactured in Syracuse, NY. In 2008 Benchmark demonstrated its strong commitment to Central New York when it opened a new and expanded R&D and manufacturing facility in Syracuse.
John kindly shared his thoughts and opinions about the design issues around high quality audio conversion;
“For almost 30 years, Benchmark has been building audio equipment for applications that demand high transparency. Some well-respected audio manufacturers specialize in building equipment that is specifically designed to alter and color the audio. Both styles of equipment are equally important in the studio. Sometimes the recording engineer is striving for a true-to-life representation of a musical source, and at other times the engineer is artistically creating a sound. In both cases, the engineer must be familiar with the sonic characteristics of the gear he is using, and must have a monitoring chain that he can trust.
“The monitoring chain must be capable of accurately conveying every sonic nuance. Noise will obscure details. Distortion and frequency response variations will color the sound. Phase response problems will alter the stereo image. Jitter and other distortion artifacts that are specific to digital audio, may add harshness to the presentation. For these reasons, the monitoring chain is usually selected for maximum transparency. Benchmark’s DAC1 D/A converters are specifically designed for transparency, and have gained tremendous popularity in the monitoring chains of recording studios and mastering rooms.
“In many ways, A/D and D/A converters are the most critical components in the digital chain. Converters can produce distortion that is very foreign to our analog world. Digital distortion mechanisms include quantization, aliasing, clock jitter, linearity errors, idle tones, and digital crosstalk. Each of these produce distortion that is inharmonic, non-musical, harsh, and unnatural. Fortunately, each form of digital distortion can be reduced to levels that are well below audibility. The required technologies have a cost, and it is not unusual to find products that take shortcuts. It is my opinion that many digital conversion products have enough digital distortion to explain audible differences.“The distortion produced by digital systems should not be confused with the harmonic distortion produced by analog circuits. In many cases, analog circuits can produce a warmth and richness that contributes to the music. There are many analog products that are specifically designed to add harmonic distortion in order to create pleasing musical effects. In contrast, the artifacts produced by digital distortion tend to detract from the music. Therefore, it is important to understand, recognize, and eliminate the distortion mechanisms that can occur in a digital system.
“Quantization errors are amplitude errors that result from mapping an analog signal into a limited set of digital numbers. 16-bit systems give us 65,536 levels to quantize the analog signal. The 16-bit quantization error (due to rounding or truncation) is 1/65,536 or -96 dB. If we added only one bit, and created a 17-bit system, we would have 131,072 quantization levels, and a quantization error of 1/131,072 or -102 dB. Each added bit doubles the available quantization levels, and reduces the quantization error by 6.02 dB. At 24-bits, the quantization error is -144 dB, and is well below audibility.
“24-bit conversion is more than adequate to eliminate quantization errors. Noise from the microphone, the room , and the analog electronics, almost always exceeds the -144 dB noise contribution of the quantization errors. The analog noise at the input of a 24-bit A/D converter is always sufficient to fully dither the quantization errors. This dithering transforms the quantization errors into a random white noise at a level that is near -144 dBFS. If we examine analog systems at the atomic level, we discover that analog systems are quantized by the movement of individual electrons. Most 24-bit A/D and D/A converters have far more noise produced by the random motion of electrons (Johnson noise) than by the 24-bit quantization. Nothing is gained by increasing the converter word length beyond 24-bits. The amplitude response of a good 24-bit converter is entirely free of quantization steps.
“Dither randomizes quantization errors so that the error signal closely resembles a random noise signal. A properly dithered digital system is the equivalent of an analog system mixed with a noise signal. 1-bit DSD systems demonstrate how effective dithering can be. Quantization noise is -6 dB in a DSD system, but with proper dithering, the 1-bit digital system can produce a continuous amplitude response that rivals that of an analog system. The quantization noise in a 1-bit system is extremely high, and must be aggressively noise-shaped to frequencies above the limits of our hearing. DSD requires re-application of aggressive noise-shaping every time the DSD signal is processed. The 1-bit DSD audio signal degrades quickly with each processing step. For this reason DSD is always converted to PCM, in DSD processing systems. Benchmark strongly recommends avoiding 1-bit and 16-bit digital systems in a production environment. Every DSP processing step adds quantizing noise to the signal. Every bit that we add to our digital system quadruples the number of DSP processes that can be performed for a given amount of quantization noise. One 16-bit DSP operation generates as much noise as 65,536 24-bit DSP operations. For this reason, all processing in the studio should be done at a high bit depth (24 or more bits, never 16 or 1).
“Digital systems must be band-limited to 1/2 of the sample rate (also known as the Nyquist frequency). If any audio exceeding the Nyquist frequency is digitized, it will be aliased (frequency shifted) to a lower frequency. Modern oversampled A/D and D/A converters only exhibit small amounts of aliasing, and all of the alias tones that occur are very close to the Nyquist frequency. Converters running at 44.1 kHz will generally show some low-amplitude alias tones in the transition band between 20 kHz and 22.05 kHz. While these tones are probably inaudible, they are uncomfortably close to frequencies that we can hear. At 2X sample rates (88.2 kHz and 96 kHz) the transition band is moved above 40 kHz, and aliasing becomes a complete non-issue. There is no additional benefit to be gained by using 4X sample rates. Most converters actually provide better distortion and noise performance at 2X sample rates than at 4X rates. 4X rates are a product of marketing departments, and offer no sonic benefits. Benchmark supports 4X rates but recommends 2X rates.“Converters can produce strange and unnatural distortion if digital codes do not have a linear mapping to analog voltage. Ideally, each digital step should represent exactly the same voltage change. Linearity errors resemble the crossover distortion that can occur in an analog class B output stage, except that the crossover errors occur at every major bit transition. Today’s sigma-delta converters have near-perfect linearity to the 24-th bit, and have a true linear response. In contrast, ladder-network converters have linearity errors that produce significant levels of intermodulation distortion (IMD). In the early days of digital audio, all converters used ladder networks, and these early converters had significant linearity errors. Some ladder-network converters hide this defect by adding a slight DC offset. Ladder network A/D and D/A converters often have very low noise and outstanding dynamic range, but measure poorly on IMD tests. Benchmark does not use ladder-network converters because we believe that IMD problems can significantly reduce listening enjoyment.
“Clock jitter can create effects similar to wow and flutter, except that it usually occurs at much higher frequencies. Jitter usually produces a non-harmonic distortion effect that is non-musical. The peak amplitude of jitter-induced distortion increases when timing accuracy decreases. Jitter phase-modulates the audio, and creates distortion tones above and below each tone in the music. In theory, a perfect, noise-free, 24-bit converter would need a sampling clock timing accuracy of about +/- 1 pSec to keep jitter-induced distortion below -144 dBFS when digitizing a full-scale 20 kHz sine wave.
“To put this into perspective, light travels 1 foot in about 1000 pSec. Perfect 24-bit audio conversion would require remarkable clock accuracy. If we wish to keep jitter-induced distortion below -100 dBFS, we need a timing accuracy of about +/- 100 ps. Benchmark converters achieve a clock timing accuracy of better than +/- 8 pSec, a level that is sufficient to keep jitter induced distortion below -130 dBFS under all operating conditions. Jitter induced distortion is well below audible levels in Benchmark converters. Some converters omit expensive jitter attenuation circuits, and may produce audible jitter-induced sidebands. Unfortunately, external master clocks do very little to improve the jitter performance of these converters. Jitter must be attenuated inside the converter, just inches away from the converter sample and hold circuit.
“Fortunately, the technology exists to reduce all of these digital artifacts to inaudible levels, leaving a conversion system with an analog response. Nevertheless, many converters fail to reach their peak performance when operating in real-world conditions. The ideal operating conditions used to generate published specifications may not reflect typical conditions in a studio. To achieve digital transparency, a converter must be capable of rejecting jitter, power line interference, and radio interference. The converter should also be capable of driving heavy output loads over a range of input and output levels. Benchmark converters are designed to maintain peak performance over a wide range of operating conditions.”
Benchmark Media Systems, Inc.
26 March 2012
You can review the entire Benchmark product range at http://www.benchmarkmedia.com