Modern PC soundcards can deliver excellent audio performance, but are sometimes held back by the computer itself. Martin Walker shows you how to squeeze out the last drop of performance.
Over the last few years, I've been monitoring the sound quality achieved by typical soundcards in my various PCs. One of the trickiest areas for soundcard manufacturers is the audio degradation that can be caused by the card picking up stray signals from the rest of the PC. This means that in a benchtest setup, a good audio specification may be achieved that's unattainable in the real world — ie. when the same soundcard is used inside your PC. For this reason, quoted soundcard specs are always likely to be 'best case' results (see the 'But That's What It Said In The Spec!' box). There are ways of minimising this degradation — keeping the soundcard well away from video cards, hard disk drives and their controllers can certainly help, and so can making sure that the soundcard is optimally set up for your signal levels.
Sadly, computer soundcards are impossible to audition until they have been installed, and so much recommendation occurs by word of mouth between owners. The Internet is a good way to come across the unexpurgated comments of users, but again, since results vary from PC from PC, even with the same soundcard, it's a case of 'the proof of the pudding...'. Even so, there are various things you can do to obtain maximum performance from any soundcard. Not all the following tips and techniques will improve your particular combination of soundcard and PC, but most of them involve little effort, and so are well worth trying.
On The Level
Many people record to a soundcard in a bit of a hit‑and‑miss way — just plugging in a signal to the line input, pushing up the line input level control until the signal peaks hovers about 0dB, and then pressing Record. However, to achieve the lowest noise with your soundcard, you need to set it up a bit more carefully. First of all, always disable (or de‑select) the mic input if possible, and failing that pull its input level control down to zero. Soundcard mic circuitry will produce far more noise inside a PC than any mic preamp in an external mixer. Apart from the line input that you will use, any other recording inputs provided (such as CD Audio or the MIDI output from the card) should also be disabled, to remove their noise contributions. If your Mixer applet (the small control panel with audio sliders) has input gain controls, then set these to unity (x1), and rely on your mixer to boost signals, since the lower the gain inside the PC, the lower the noise is likely to be as well.
Since a soundcard is effectively a digital recorder, the best way to optimise levels should be similar to what it is for DAT recorders — selecting a normal operating level to give you an amount of headroom suitable for the type of music you are recording (see 'Setting DAT Record Levels' in the January '95 issue for more details). Unfortunately, you can't always send the full output of a mixer into the line input of the soundcard, since many cards will be designed for the lower 'consumer' level of ‑10dBv. The power supply voltages on the soundcard are much lower than those of typical mixers, so sending 'healthy' mixer output levels to the soundcard is likely to overload the initial stage of the soundcard, however low you set its input level control. If you go to the other extreme, and turn the soundcard input level control to maximum, reducing the input signal level to compensate, you'll probably find that the background noise from the soundcard circuitry goes up a bit as well.
To optimise recording levels, you'll need to send a few different levels to the soundcard, with its input level adjusted in each case so that the PC shows as close to 0dB recording level as you can set, without ever running into clipping. Most modern WAV editors, including Steinberg's WaveLab and Sonic Foundry's Sound Forge, have accurate metering systems that will help in setting levels, rather than a pretty simulated LED display without numbered markings. Record a few seconds of a 1kHz sine wave (either using the line‑up oscillator provided on many mixers, or a synth playing B natural two octaves above middle C). Start with 0VU showing on the mixer meter, record a few seconds of this tone at just under 0dB on your soundcard level meter, and then play it back, listening for distortion. If this sounds OK, increase the mixer to +3VU, then +6VU, reducing the input level control on your soundcard in each case, and then listen again. Using a small Spirit Folio desk with the mixer output meter reading +6VU, I found there was fairly obvious distortion from an AWE32 on the sine wave when I played the soundcard recording (despite the soundcard's meter never running into clipping). I found that I could get as high as +3dB VU (showing on the mixer meter). Since the 0dB VU reading on most mixers emerges at +4dBu from the output sockets, this output level is actually +7dBu, or about 1.75V RMS (about 5V peak to peak).
Despite having a 120dB capability for the digital output, if you are recording using the soundcard's A/D converters, and these give 80dB signal/noise ratio, then this is your overall system figure.
Once you've found this level, and set the soundcard input level control to suit, you'll have probably got the optimum setting from the noise point of view. For playback, you'll probably get the lowest sample playback noise when only the wave volume control is pushed up. Any on‑board MIDI units, such as wavetable synths (or, heaven forbid, two‑operator FM synths!) should have their mixer levels pulled down, unless you need their musical contributions in the mix.
These procedures may improve the noise by 10dB or more over your current settings, depending on the design of the card. Certainly, AWE32 and Gravis Ultrasound cards showed a significant improvement, although the AWE64 Gold seems to have its input level control before the rest of the circuitry, as its noise stayed exactly the same, whatever the setting of its input level control. In this case, higher mixer levels can be easily achieved without overloading the card, and this does make it easier to line up with a mixer.
Getting Wired
Another area in which people often experience problems is when they wire up the output from their soundcard to an external mixer. Since the PC is already earthed to the mains supply, an additional earth path via the input of a mixer produces an earth loop — as soon as you plug in the soundcard, you'll hear a noticeable hum. Unfortunately, due to the huge number of digital signals flying around inside the PC, you are also likely to hear other sounds (admittedly at fairly low levels), such as warbling, whines, or regular ticking sounds, mostly associated with processor or hard disk activity. In some cases, the background sound will be as if every one of the notes is sounding simultaneously, a bit like an electronic beehive. You'll have to accept some of these sounds, unless you buy a more expensive soundcard that features more comprehensive shielding. However, much of this grunge may be the result of the earth loop, and can be greatly reduced if you have a mixer with balanced line inputs of the TRS (Tip Ring Sleeve) variety. Many modern mixers feature these, from the tiny Spirit Folio range upwards, and by making up a special lead from PC to mixer, you can remove the effects of the earth loop. In essence, you need to make up an unbalanced to balanced cable, so that first, the earth loop is broken, and second, any interference picked up along the cable itself is cancelled out by the balanced connection at the mixer end.
You need to make up a special cable for each soundcard output channel, using 2‑core plus screen cable (the sort sold for balanced mic use is fine). At the mixer end, each channel will need a stereo jack plug, with one core of the cable connected to the tip, and the other to the ring, with the screen attached to the jack sleeve as normal. Sadly, most soundcard manufacturers use a 3.5mm stereo jack socket for their line output, which makes wiring the other end of the cable far more fiddly. I've just installed an AWE64 Gold card in my machine for general purpose use, and this features a pair of gold‑plated phono sockets, which is much easier to deal with.
The secret is to attach the 'live' side of each soundcard output channel to the core of the cable connected to the tip connection at the mixer end, and the 'earth' side of each soundcard connection to the cable core attached to the ring at the mixer end. This provides the balanced part, and since the screen connection from the cable is already connected at the mixer end, the entire cable is screened, although it is preferable to connect a low resistor value such as 50Ω between the soundcard end of the cable screen and the PC earth, so that you don't get an almighty hum if any other connection comes adrift. A 50Ω resistor is quite sufficient to stop the earth‑loop hum, and since this pseudo‑balanced connection is now effectively disconnected from the actual PC earth, your signal should be much cleaner. On my machine, these leads made a huge difference to the level of grunge emerging with the soundcard signal.
An Old Buffer
Once you've solved the hums and hisses associated with the audio circuitry, the other audible problem that may affect you is clicks. These come in several forms, but are nearly always caused by small unwanted gaps in the audio waveform. They can occur during the recording process, or only when playing back previously recorded data. If a click always occurs at the same moment during playback of a file, then it is likely to have been permanently recorded as part of the file. Use the Zoom function of your editor to view the waveform in detail, and you will probably be able to see the click. Many waveform editors will allow you to remove it by 'drawing in' the waveform at this point, and if it's a stray one‑off anomaly, this is the easiest solution.
The ultimate solution is to give up — not to sell your computer and buy a guitar, but to remove the audio circuitry from the inside of the PC altogether.
For consistent problems during record or playback, the problem could, unfortunately, be caused by many things — a slow hard drive, badly written Windows soundcard or graphics drivers, or problems with full duplex operation (recording simultaneously with playback of previously recorded material). Many of the timing‑related problems can be helped by intelligent use of software buffers. These simply store up a portion of the audio signal in advance, so that if any component of your PC decides to do something else for a moment (such as the hard drive lurching to a completely different spot on the drive to continue reading a fragmented file), there's enough prepared audio in the buffer to allow it to 'catch up' before an audible gap or stuttering occurs. The size of these buffers is very dependant on the software, your soundcard, hard drive and so on, and most applications that feature audio recording and playback will have various options for their buffers. Since each application has its own ways of dealing with such problems, you should refer to the manual and help file for further information if you run into problems, although, in general, with a reasonably powerful machine, the default settings will probably be fine.
On the hardware side, most soundcards still use DMA (Direct Memory Access) to move audio signals to and from the hard drive, although this does tie up a significant amount of your processor time. Each design of card will require a certain size of buffer for this DMA, to optimise the flow of data. Most modern software will provide a way to set up the correct DMA buffer size automatically — Cakewalk uses a 'Wave Profiler' to determine the settings required, and Steinberg's Cubase provides the 'Detect DMA Blocksize'. Thankfully, these adjustments only need to be made once, unless you install a new soundcard.
Going Digital
Since most soundcard audio problems are caused by audio signals inside the PC, one way to side‑step the issue is to go digital — use external A/D (analogue to digital) and D/A (digital back to analogue) converters, and leave only the digital signals inside the PC. The cheapest option is to buy an AWE64 Gold card, which comes complete with an S/PDIF (Sony Philips Digital InterFace) socket. While this allows only digital output, any pre‑recorded WAV files, such as those from CD‑ROMs, or created by software synthesis, will emerge in full 16‑bit digital glory, and a potential 96dB signal/noise ratio. To hear the improvement, you need to plug this digital Out into a co‑axial digital Input. Most DAT recorders feature these, although those that only have the optical type will be unsuitable. Once it's all connected up, switch the input of your DAT machine from the normal Analogue to 'Co‑axial' or 'Digital'. Then enter Record‑Pause mode (most models will let you do this without having a tape inserted). You'll then be able to monitor the output of the D/A converters via the normal DAT output sockets.
You can also buy stand‑alone D/A converters, and although there are rackmounting devices available primarily for studio use, with amazing audio specifications, the most cost‑effective solution for stereo is to buy a mass‑market hi‑fi type (see my 'Hi‑Fi Tweaks' feature in the August issue for more details on this).
To get further improvements, you need a soundcard digital input as well, so that you can record using external A/D converters too. More and more soundcards are becoming available with full digital I/O, although these are normally considerably more expensive than consumer types. This looks set to change — just before I finished writing this, a press release from Et Cetera informed me that the new Maxisound Home Studio Pro 64 card was available. Although I haven't yet seen it in the flesh, it combines a full duplex stereo soundcard (with all the usual frills and extras) and both S/PDIF input and output, at the very reasonable price of £249.
The Future
There's a whole clutch of multi‑channel soundcards appearing, and SOS hopes to be reviewing some more of these over the next few months. The main thing to watch out for is the driver implementation. If you're using a MIDI+Audio package, such as Cakewalk, Cubase or Emagic Logic, you can use any stereo soundcard with standard drivers, and the eight or more audio channels are mixed in real time to emerge through this single stereo output. If you buy a soundcard with four hardware channels or more, the most efficient way to implement the driver is as a special multitrack driver. However, unless the sequencing package specifically supports this special hardware driver, you may still only be able to access the first two channels.
For instance, Cubase VST for Windows supports multi‑channel cards such as (among others) the Korg 1212 as well as the forthcoming cards from Event Electronics; whereas Emagic's Audiowerk8 card (which works very nicely with Logic Audio) currently still needs a special driver to be used fully with Cubase VST, and PC Cubase users can currently only access two Ins and two Outs. As both software and hardware get more complex, and achieve more and more at lower and lower prices, these initial incompatibilities are bound to occur.
Musicians are demanding more and more hardware and software channels. The main reason for increasing the number of hardware channels has always been to enable EQ and effects to be added to each sound separately. As real‑time EQ and effects are already available from Cakewalk Pro Audio v6, and Cubase VST v3.5 for PC, and from the new version 3.0 of Logic Audio (expected to be shipping by the time you read this in mid‑October), I suspect that many people may find themselves content with fewer hardware channels than they expected. After all, if you can EQ and add effects to eight or more software channels (assuming that your PC has enough processing power), you effectively have mix automation. With four or eight hardware channels, I can see many musicians using two channels for the main stereo mix, and the rest as effect sends to an outboard rack, rather than separating each track in the traditional manner.
Although many people are fully expecting more power for less money, it seems likely that, as software packages achieve what until recently would have needed a bank of extra hardware, the technical 'carrots' dangled before us will result in our spending more overall, rather than less. PCs will need to be upgraded more regularly to have enough power to achieve what the software promises, and the number of hard drives and their sizes will also expand significantly, simply because applications like Cubase VST for Windows will open our eyes to yet more tantalising possibilities. And, of course, the bigger the hard drives, the more there will be to back up. C'est la vie!
Real‑World Figures
PC soundcards are in a hostile environment — they sit in a sea of digital signals which inevitably degrade performance to some extent. For this reason, the theoretical figures published by manufacturers are unlikely to be achieved in practice. This is not anybody's fault, just a simple fact of life. In the real world, soundcard noise figures will vary to some extent when measured in different PCs.
To give you an idea of real‑world figures, I measured s/n ratios in my machine with several soundcards. Once the soundcard has had its levels and mixer settings optimised (see the first part of this article), all you need to do is enter Record Monitor mode in your WAV file editor, so that the level meter is flickering at a low level in the absence of a music signal. This level (relative to the maximum 0dB level before clipping) is the s/n ratio of the recording side of the card in your machine. If you have any options in your editor for the automatic calibration of DC offset, make sure these are active first. WaveLab seems to do this automatically, but in Sound Forge, you click on the DC Adjust box to remove any constant offset automatically during recording. Press the Calibrate button to set this up, before reading the signal/noise ratio.
With an AWE32 Plug & Play soundcard, the meter hovered at about ‑68dB in each channel. This then is the signal‑to‑noise ratio for the card (in my PC). With an AWE64 Gold, the noise dropped slightly, to ‑70dB. Both of these figures are perfectly adequate for general‑purpose recording, and a lot better than with the microphone input left on (this measured about ‑61dB on the AWE64 Gold). To put these figures into perspective, an elderly Gravis Ultrasound card measured ‑55dB, a Sound Galaxy Basic 16 card measured ‑57dB, and the MaxiSound 64 (reviewed in the February '97 issue) measured ‑72dB.
Creative Labs did expect a rather better measurement than this for their Gold card, and were very helpful, insisting on lending me another Gold card to try. However, this measured ‑65dB (in my system), which rather suggests that there may be significant variation even between items from the same manufacturer. Indeed, a recent report in a PC magazine measured both AWE Gold and Gravis Ultrasound Plug & Play Pro at ‑65dB, but the Guillemot MaxiSound 64 at ‑42dB! Luckily, I visited Paul White a couple of days before finishing this feature, and took the opportunity to measure his AWE64 Gold card as well. The one in his machine measured a very healthy ‑80dB, which is much more in line with Creative Labs' own typical measurements, but I don't yet know whether it was his card or his PC that improved things. It just goes to show how variable the real‑world results are, and why it's impossible to choose a soundcard solely on published specs.
But That's What It Said In The Spec!
Let's lay one myth to rest once and for all. If you see the words 'CD quality' associated with any product, all this means is that it uses 16 bits of information at a sample rate of 44.1kHz. This doesn't automatically give good audio quality, and indeed these two words are quite often quoted instead of a proper spec. I do appreciate the problems that face soundcard manufacturers in particular, since test figures may vary significantly from machine to machine, so published figures can be either 'best case' (possibly achievable only in benchtest conditions) or 'typical'. Marketing departments are notorious for 'embroidering' the facts — a claimed 90dB s/n ratio for a soundcard may become 70‑80dB in the real world, although it may still be possible that 90dB is occasionally reached.
Now that digital I/O is appearing on higher‑end soundcards, very impressive figures for noise are starting to be quoted. It's important to keep these in perspective. If you see 120dB quoted for an S/PDIF digital output, this is the theoretical 20‑bit value (20 x 6 = 120). You will always get a significant improvement in fidelity by using such outputs, because you're bypassing the soundcard's D/A converters. If you use the digital output, and plug into a higher‑quality external converter (in your DAT recorder, for instance) your signals are likely to sound cleaner and quieter, but the actual noise levels will be determined by the weakest link in the chain. Despite the 120dB capability for the digital output, if you are recording using the soundcard's A/D converters, and these give 80dB signal/noise ratio, then this is your overall system figure.
Another thing to watch out for is what exactly is included in the circuit to achieve a particular figure. For instance, a soundcard frequency response of 15Hz to 50kHz (±1dB) sounds very good, and indeed it is, but with a sampling frequency of 44.1kHz, the top‑end response cannot theoretically go above about 20kHz. In this case, the quoted figure is probably for the electronics alone, without sampling being taken into account. Again, the figure itself is accurate, but can be misleading if you take it out of context.
The ART Of Noise
The amount of noise in any electronic circuitry is measured relative to the highest normal signal level, and the result is normally expressed in decibels. Positive figures (for example, 80dB signal‑to‑noise ratio, or s/n) mean that the signal level is higher than the noise — 80dB higher, in this case. If the figure is negative, this is because it is referring to the noise: ‑80dB would mean that the noise is 80dB lower than the signal. Each time the ratio changes by a factor of two, the decibel figure alters by 6dB. If, for example, a signal level alters from 1V to 2V, this is a +6dB change. If, on the other hand, it alters from 1V to 0.5V, this change is ‑6dB.
To give a few examples, the theoretical s/n ratio for an 8‑bit sample is 48dB, since each of the eight bits contributes 6dB to the overall figure (8 x 6 = 48dB). A 16‑bit sample has a theoretical s/n ratio of 96dB (16 x 6 = 96). Typical noise levels for consumer DAT recorders are about ‑85dB (or, expressed the other way, a signal‑to‑noise ratio of 85dB), whereas more expensive studio models are likely to be 90dB or more. The difference between the theoretical and real‑world figures is due to electronic design. Anything better than 80dB is likely to be good enough for general‑purpose music recording as long as you're careful to optimise recording levels. In the case of typical rock music, with a narrower dynamic range, this noise contribution will be largely inaudible in context. Many people are happy recording using soundcards that only measure about 70dB — this is, after all, the sort of figure achieved by cassette‑based recorders with Dolby B noise reduction, and a typical professional reel‑to‑reel recorder running without noise reduction.
Going Upmarket
Upmarket soundcards can introduce more expensive shielding, which helps keep the hostile digital environment away from the sensitive audio circuitry. It might even be feasible to totally enclose a soundcard inside a metal case, with only the edge connector protruding. However, even then, stray signals might get in through the power supply or data lines, and you'd probably need an empty card slot on either side of the soundcard to fit it in! The ultimate solution is to give up — not to sell your computer and buy a guitar (although I secretly suspect that many musicians might be sorely tempted), but to remove the audio circuitry from the inside of the PC altogether. Packages are appearing with the 'central nervous system' on a card inside the PC, and all the converters (which change the digital signals to and from analogue) in external boxes. Published audio specs are then much more likely to be 'just what it says on the box'.
Once the audio circuitry is placed in its own little box, the digital interface left inside the PC can get on with what it does best — moving and manipulating large quantities of digital data. With the slow but sure movement from the ISA buss (8MHz) over to the PCI one (normally 33MHz), the possibilities keep on growing, as more data can be moved at a faster rate than before. In addition, if some intelligence is put on board the digital interface, it's possible for the computer's main processor to send a command, and then leave the digital card to carry on by itself. This is how 'bus Mastering' works with hard drives, and its inclusion in many SCSI devices tended in the past to make them faster than EIDE ones. Placing some intelligence, in the form of a DSP (Digital Signal Processor) chip, on a soundcard can give it a similar performance boost — for instance, the Turtle Beach Fiji and Pinnacle cards both use an industry‑standard Motorola 56002 DSP chip, and this gives them lots of extra processing power.
Although 'bus Mastering' audio removes much of the workload from your computer's main processor when you're shifting digital audio signals about, a DSP can provide far more capabilities than this. You can also let it get on with lots of other goodies such as EQ and effects, while the main computer processor concentrates on running the remainder of your software. This can result in a much more open‑ended system, allowing you to expand in several directions, as your needs and budget dictate. For an example of such a external DSP‑based system, take a look at my review of soundscapeshdr1.html.