Winter Audio Reference: On-Board, Consumer, and Pro Solutionsby Derek Wilson on February 3, 2005 12:05 AM EST
The Anatomy of a Sound Review (Electrical Analysis)
Sound hardware is one of the most straight-forward, and yet difficult, components of a computer system to review. The human aural system is ill-equipped compared even to the eyes when it comes to judging subjectively the "goodness" of one audio system versus another, especially when there is any time between the tests. Of course, this doesn't mean that listening to a high quality consumer electronics audio system isn't head and shoulders beyond listening to analog audio from a PC based system.
Running digital audio streams from a PC into a receiver or speaker system will usually be higher quality than running analog audio (depending on the DAC), but such a vast majority of PC users either run PC speakers, headphones, analog surround systems, or some other such setup that the DAC (digital to analog converter) on the sound card becomes hugely important factor in the overall sound quality. Also key is opamp (operational amplifier) selection. Opamps are transistor-based analog components that can be used for a number of purposes. In the case of the sound card, opamps take the output analog signal from the DAC, and both isolate the input from the output while amplifying the signal to a proper level. Opamps can also be used for level control. The way that the circuit is built would be different from vendor to vendor, but the way sound output is designed lends itself to viewing maximum output level as 0dB and everything else as an attenuated signal. Testing of (and indeed listening to) analog sound output is better done at full scale (0dB - the maximum output level of the system as an input to the rest of the chain).
Electrically, the components are easily testable (provided you have no unforeseen cabling or interference issues). Ideally, we would use a signal generator, a spectrum analyzer, and other high end equipment to test each audio card. Unfortunately, this equipment is expensive, of limited use other than in sound testing (to us), and just not worth it at this point. RightMark Audio Analyzer is an excellent stop gap solution to expensive measurement device route. RMAA allows us to plug the output of one card into the input of itself or another and analyze the signal. This is excellent, except for the fact that it depends on the ability of cards to do high quality recording as well as playback. Consumer level cards are targeted at higher quality playback than recording, and a loopback test is really a composite test of both. As long as we consider this going into the situation, we won't have any problems.
Of course, the glitchy cable or bit of interference is a curse that doesn't go away. One combatant to interference when running analog audio is to use balanced signaling. This is a method by which one audio channel is communicated via two signals and a ground. The noise or interference introduced between the source and destination can be kept to a minimum. While musicians who record using analog equipment and audiophiles who listen to music without going digital will already know and love balanced signaling (while loathing cable prices), consumer audio cards do not enable balanced connections. The standard 1/8 th inch stereo mini jack is not the best connection for audio, falling short to stereo RCA. Quarter inch is better for analog, especially of the balanced TRS variety. Balanced XLR cables are the preferred audio connection for musicians who use analog recording equipment.
There are a few select pieces that we are testing when looking at analog audio quality on a PC. In sending sound out to speakers, the card must convert audio to analog using a DAC, then prepare this signal for output using opamp circuits. On the input side, the ADC (analog to digital converter) is tested as well. The metrics for these tests are explained here:
Frequency Response - This has to do with a system's ability to reproduce sinusoidal signals at different frequencies. In our tests, we will vary our signal from 20 Hz to 20 kHz and measure the amplitude of the received signal. The plot that we will see from RMAA is a plot of amplitude in dB below FS (fullscale) versus frequency. Ideally, we would like to see a flat line indicating no change in amplitude over frequency, but invariably, there will be some drop off at higher and lower frequencies, especially at sampling rates lower than 96 kHz. This measure is very dependant on opamp quality being very high.
Noise - The noise measurement here will represent the ambient noise present on the line when nothing is being played or received. The limiting factor on this is a combination of the DAC and the analog circuitry. Just as with Dynamic Range, the number of bits used to encode audio really dictates minimum noise levels. With 16bit audio, when looking at dB output, it is only able to generate a minimum of -96dB below full scale, so it is obvious that any noise that gets introduced into the system (any bit twiddling) will increase the dB level of the noise to above the theoretical limit of -96dB.
Dynamic Range - Recently introduced in the graphics industry, the push for higher dynamic range is a push to increase the difference between the largest and smallest reproducible values in a given data set. In the case of audio, dynamic range signifies the effective range of amplitudes that the hardware can reproduce. This range, as we will report it, is in A-weighted dB, and RightMark simply measures the difference between the peak output level of a 1 kHz signal and the noise floor. The theoretical limit on maximum dynamic range (and thus, minimum noise) comes from the fact that every 6dB increase in signal represents a doubling in power output. As we use bits to represent linear increases in power, 16bits represtents 16 "doublings" in power or 16 * 6 dB = 96dB in maximum dynamic range. The dynamic range (and noise level) of a 16 bit system is generally limited by the bit width of the data. In a 24 bit wide setup (such as is used on DVD audio and SACD formats), the maximum on dynamic range is 144dB, and this is limited by analog circuitry. The voltage levels that would be needed to drive a signal at low enough levels to extract the maximum dynamic range from a 24bit system on current ADCs is way too low to be feasible. It may be that with some sort of biasing and tweaking, it would be possible to get at these extremely subtle changes in sound. Of course, this is not currently practical or necessary.
Total Harmonic Distortion (THD) - When a tone is generated at a frequency (called the fundamental for this test) on a non-linear device such as an amplifier, tones are also found at the harmonics (integer multiples of the fundamental frequency) at lower levels than the fundamental. To find THD, the dB level of harmonics is added together and total output power is looked at as a percent of the fundamental's power. In actual music, this happens for all frequencies played, so having THD as low as possible is desirable.
Intermodulation Distortion (IMD) - When two sinusoids are generated at the same time (at frequencies F1 and F2), they produce intermodulation distortion. This comes in the form of attenuated tones at frequencies: F1+F2, F2-F1, 2F1+F2, 2F2+F1, 2F1-F2, 2F2-F1, etc. This continues, but at decreasing levels per intermod. Also, if any of these values are larger than the nyquist (samplerate/2), they will wrap around from DC. We look at the total power of the intermods as a percentage of the total power of the signals at F1 and F2 (for RMAA's test, F1 is at 5dB below FS, and F2 is at 17dB below FS). So, now that we know what this is, we need to tackle what this means. Let's say that there are two high frequency signals separated by 1 kHz. IMD could easily muddy up both the mid-range and low end at the same time (since distortion wraps around after passing the nyquist). Low end intermods will also affect the whole frequency range, but in order to really get an idea of IMD performance, multiple frequencies must be tested (which RightMark now has an option of doing). Frequencies with a 1 kHz separation are tested across the audible range.
Stereo Crosstalk - This is the amount of signal that bleeds through from one channel to the other at a given frequency as measured in dB. Obviously, the lower, the better, and the result of not seeing good measurements here is that the signal which should come through on one channel will show up a little on the other.
In the end, the speaker is usually the weakest link of the audio chain. Of course, introducing a minimum number of problems along the way definitely helps, and we won't pretend that it's a good idea to go with inferior audio hardware. We also won't tout the glory of owning the most expensive sound card on the market for its analog listening potential. The human ear is, again, not the most highly tuned instrument. With training, it is possible to understand and recognize the differences between a high quality audio system and an average one. Listening to an average system and then listening to the same audio on a high quality system immediately thereafter should make it apparent to everyone that there is a difference. Most people will just say that the well-built system will sound "better" and leave it at that. Audiophiles may say that the average system sounds "colored" or talk about hearing artifacts.
It is not possible to tell when a note wasn't loud enough because of poor frequency response, to hear a harmonic tone at certain frequency, and to notice an intermodulation distortion at another. Music, and indeed most audio (with the exception of some electronica and a few choice academic projects), is continuous. Our tests are based on at most two sine waves played at the same time. There is distortion everywhere in all audio no matter what you do, and controlling it toward a desired state for a specific application is the key in audio system design.
Moving on from the technical side of audio analysis, we look at the end user experience.