If you're designing the analog portions of audio systems, it helps to know two things: what your listeners expect, and how you're going to objectively demonstrate that you're delivering on those expectations. Yet it's difficult to relate engineering measurements to the subjective experience of hearing. The idea that hearing isn't a linear continuum started to crystallize for me in June.
National Semiconductor wanted to show me some products it had optimized for the absolutely purest consumers of audio systems-the hardcore, two-speaker audiophile. These products comprised a family of high-voltage op amps with extremely low distortion and an audio power-amplifier driver. Both are designed to replace large handfuls of discrete components in top-end audiophile pre-amps and amplifiers. Designers of ne plus ultra audio systems still insist on discretes, and National has set out to change their minds.
National set up an acoustically ideal sound room with a pair of Wilson Audio Specialties Watt-Puppy speakers and a reference-design analog audio chain built around the new chips. It also had its own CDs to demonstrate characteristics that true believers listen for, but I was invited to bring my own disc, too. So, I grabbed a CD of Beethoven piano sonatas from the carrier on my car's sun-visor. The grand piano is a subtle instrument, and Beethoven piano sonatas are unlikely to undergo much in the way of post-production "enhancements."
My expectations were low, because my hearing has been attenuated by age, automobiles, airplanes, and motorcycles. Knowing that, I walked into that sound room thinking its $26,000 worth of speakers connected to National's custom electronics was all overkill for my $50 ears. That is, I figured that $50 was all it was worth spending on headphones or speakers for my personal use. The upshot? I was shocked at what I heard.
On National's demo recordings, I could tell which cymbals rang longest. I also could hear the singer's lips make that sound when you separate your lips and there's some saliva on them. On my Beethoven CD, I could hear the pedal work-not the mechanical action, but the difference in tone as the artist's foot released-sustain exactly on the beat.
I was puzzled. I have the furthest thing from "golden ears" that one can have without wearing prostheses. How could I be getting so much new experience from a recording I had listened to hundreds of times? I put that question to Mark Brasfield, principal audio applications engineer at National and creator of its sound room (Fig. 1).
Interestingly, Brasfield confessed that he also has some hearing loss. Yet he said industry studies revealed that, up to a point, people with reduced hearing acuity actually get more out of high-end audio than people with perfect hearing.
To get an idea of what kinds of tests you could run to compare audio systems, I spoke to Bruce Hofer, founder and president of test-equipment maker Audio Precision (see "The Challenges Of Audio Testing" at www.electronicdesign. com, Drill Deeper 16803). Audio Precision is probably the most respected maker of purely audio test gear (Fig. 2). But if I was hoping for unambiguous metrics, Hofer discouraged me.
"After three decades now, one realizes that you don't understand what you don't understand about the ear and the brain system," Hofer said. "At Audio Precision, we're all about making objective measurements on audio devices. But the world is also about perception and measurements, which, by their very nature, are not going to track perfectly with that which you're really trying to determine."
Hofer said that Audio Precision doesn't have an audio distortion analyzer that says "Great/Good/Bad" yet. His test gear simply provides numbers, measures of performance, usually cut along different axes like frequency response, distortion factor, or noise. While these metrics all are valid, Hofer said, the ear's behavior is more subtle than that.
The conversation then veered into the broader topic of the way companies and researchers use human subjects, rather than precision test gear, in their evaluations. Hofer said that the best examples include Fraunhofer and Dolby, which use human subjects listening to real recordings to gauge how well their lossy compression algorithms capture real performances (see "Testing For Audio Transparency," Drill Deeper 16802).
The analytical measurements that Audio Precision analyzers make are meant to avoid the downfalls of subjective measurements by describing signals that are highly defined mathematically, explained Hofer. It's a straightforward process to describe the effect of distortion products, hum products, and noise products, with respect to an original mathematically defined signal.
"But even then, if you digitally recorded a broad multitone input, the measurement will show some tones, not attenuated, but entirely missing," said Hofer. "The lossy compression algorithm made the decision not to waste any bits on certain parts of the signal because those parts didn't meet the threshold for audibility."
And if the algorithms are truly optimized, one's ears and brain won't notice the missing information, even though an analog test instrument will.
Continued on Page 2...
So if designers can't nail down the objective measurements that subjective experience relates to, what should they do? Apparently, most companies concentrate on power efficiency and integration level.
I spoke with a number of National's competitors in the analog audio chip market, and none of them are pursuing the extreme audiophile market. Instead, they're focusing on cell phones, personal media players, and home theaters. Most said that direct comparisons of listening experiences are impossible.
When it comes to earbud/headphone-oriented devices such as MP3 and AAC players, they say, such comparisons are especially impossible. That's in part due to the contribution of the actual acoustic transducers next to or in one's ears, which is significantly higher than any electronic component in the signal chain.
It's also impossible partly because the digital signal is heavily processed to fool the ear in various ways. For example, digital processing tries to make you think the sound source you're hearing is somewhere in front of you, rather than in the middle of your head, which is what you would hear with a pure stereo signal. In fact, Fraunhofer and Dolby algorithms can, to a considerable extent, simulate a complete surroundsound experience with a decent set of earphones or high-end earbuds.
Taking that into account, what those other chip companies have been up to is represented by their newest analog-audio product releases. (I'm including class D amplifiers as part of the "analog" category.)
Texas Instruments launched the largest flight of new chips, many of which are class D power amps. For example, TI is now delivering its DRV600 stereo 600-Ω line drivers, which eliminate the need for coupling capacitors on inputs and outputs.
For driving speakers, rather than 600-Ω lines, TI's TAS5162 stereo digital amplifier power stage can drive a 6-Ω bridge-tied load (BTL) at up to 210 W per channel with 10% total harmonic distortion. Efficiency is greater than 90%. For maximum dynamic range, it runs from separate 12- and 50- V power supplies. For lower output-level requirements, TI's TAS5176 can drive six channels at 15 W each or three channels at 30 W. Both chips require only simple LC output filters to remove class D pulse-modulation artifacts.
TI's TAS5414 and TAS5424 fourchannel digital audio amplifiers target use in automotive head units and external amplifier modules. Both provide four channels at 23 W continuous into 4-Ω or 30 W into 2-Ω speakers at less than 1% THD+N. The difference between them is that the TAS5414 has single-ended inputs, while the TAS5424 is differential.
Further up the signal chain, integration is the name of the game. TI's TLV320AIC3101 for digital cameras is a low-power stereo audio codec with stereo headphone amplifier, a digitally controlled stereo microphone preamplifier, and automatic gain control (AGC), with mix/mux capability among the multiple analog inputs.
The chip's programmable filters can remove zoom-motor noise. The playback path includes mix/mux capability from the stereo digital-to-analog converter (DAC) and selected inputs, through programmable volume controls, to the various outputs.
Analog Devices also has been active in class D amps. The ADAU1590 and ADAU1592 are two-channel BTL power amplifiers with a sigma-delta modulator to drive the pulse modulation. This allows for a microcontroller interface to control reset, mute, and programmable gain amplifier (PGA) gain, as well as outputs for fault-reporting signals. For use with a separate modulator, the ADAU1513 is a basic two-channel power stage.
Maxim Integrated Products offers speaker-driver entries, but they aren't class D devices. Instead, Maxim introduced two class G speaker amps, the MAX9730 and the MAX9788. The former is general-purpose; the latter is optimized for driving ceramic speakers.
Class G amps have a push-pull stage like class A-B, but they add a second higher-voltage power supply that only kicks in when signal peaks rise above a preset level. In the MAX9730's case, it will drive 2.4 W into an 8-Ω load from a 3.3-V supply. Piezo speakers are different, because they need a big voltage swing to achieve enough deflection to move enough air to make a loud noise. The chip's charge pump can supply greater than 700 mA of peak output current at 5.5 V dc, guaranteeing an output to the piezo speaker of 14 V p-p.
Continued on Page 3...
For digital-camera audio recording, Maxim announced the MAX9814 microphone amplifier with automatic gain control (AGC) and low-noise microphone bias. Integrated AGC lets designers optimize the signal level prior to digital signal processing. The device also integrates a low-noise preamplifier, variable gain amplifier (VGA), output amplifier, and an internal low-noise electret- microphone bias-voltage generator.
Back with the audiophiles
National's demonstration centered around two products. The first was a pair of audio op-amp families with typical THD+N of 0.00003% (guaranteed maximum is 0.00009%). Other performance specs include a 2.7-nV/√Hz input-noise density, a 60-Hz 1/f noise corner, 20-V/μs slew rate, and a 55-MHz gain-bandwidth product.
There are two package variants of the 44-V LME49860 dual op amp and single, quad, and dual versions of the 34-V LME4978x0. The nominal ±22-V LME49860 is unity-gain stable down to ±2.5 V. Over this supply range, the amplifier maintains common-mode rejection ratio (CMRR) and power-supply rejection ratio (PSRR) better than 120 dB and a typical input bias current of 10 nA.
On the input, the LME49860 can swing to within 1 V of either powersupply rail when driving 2-kΩ loads or to within 1.5 V when driving 600-Ω loads. The LME49710, LME49720, and LME49740 amplifiers have lower operating voltages and similar specs.
The other new product, the LME49810, is a monolithic 200-V audio power amplifier driver with an integrated Baker clamp. Like the op amps, each LME49810 can replace a couple of dozen hand-selected and matched discretes in a high-end audiophile system. The LME49810's function is to drive high-power discrete output bipolars with up to 50 mA, for systems delivering up to 3 kW. When implemented in a complete power amplifier design, typical THD+N is 0.0007%. Other specs include a 50- V/μs slew rate and a 110-dB PSRR.
That Baker clamp deals with input signal peaks. It's implemented by means of an array of diodes connected between the base and collector of a bipolar transistor. It also prevents the collector-emitter junction from saturating, making clipping less noticeable by removing the high-frequency spike that occurs when the transistor recovers from saturation.
In National's sound room, all of the electronics except for the CD player are custom. The op amps are used in the
DAC signal-path (four in each stereo channel), as well as in the power-supply regulators. The audio power amplifier drivers are, obviously, in the power amps. How much would a commercial implementation cost?
Brasfield estimates that the current National prototypes could be sold for approximately $300 each, bringing the cost for the electronic portion of the signal chain to a few thousand dollars. That's exclusive of the speakers. Unfortunately, Brasfield said that Wilson Audio Specialties has informed him that those $26,000-a-pair Watt Puppies have been obsoleted and their replacements will cost a whole lot more. Talk about "golden ears."