Reading a review of some digital delay unit can often be confusing if you're unfamiliar with the jargon used, so Paul Williams sets out below to lay bare the technology and terminology of today's common delay-based studio effects.

Time delays of audible sounds are a naturally occurring phenomenon since the sound waves take a finite time to travel any given distance; 3 milliseconds (3 thousandths of a second) for every metre to be precise. The effect is demonstrated well at an open air sporting event, for example, where the sound from the various PA speakers meets the listener at slightly different times resulting in a confusing multiplicity of apparently separate voices.

One of the few applications which makes use of this natural phenomenon for recording purposes is in reverberation chambers or 'ambience' rooms. These, in common with churches and unfurnished halls have hard walls which reflect sound well, with little absorption. A loudspeaker when suitably placed in the room will deliver its sound so that the walls, floor and ceiling bounce it back and forth and all around, causing echoes, which become increasingly frequent and diffused, until they finally decay away to silence after 3 seconds or so (depending on the absorbing properties of the walls). A microphone is then used to pick up the sound as it crosses the room from one surface to another.

The early pioneers of audio soon realised the potential of a device which could artificially create audio delays to produce natural, as well as some decidedly unnatural sound effects. To create a delay, it is necessary to store the incoming sound somewhere during the delay period, after which the sound can be retrieved and replayed.

Tape Echo

Magnetic tape was one of the first storage media to be used for this purpose, being inexpensive, versatile and fairly easy to implement, given a tape recorder with a third head fitted downtape. A signal then recorded via the normal record head would be taken downtape which, after a predetermined delay dependant on the distance between the heads and the tape speed, would reach and be reproduced by the extra head. Some of this signal could then be mixed into the record circuit again, to produce further repeats of the signal, until the sound decays away to silence (or rather tape noise!). This sounds rather familiar doesn't it? Well not reverb actually, but echo. Figure 1 makes the arrangement clear. Perhaps one of the most popular effects units ever made was the WEM 'Copicat' echo machine based on the principle described above.

Tape echo machines however, are not, without fault; they introduce tape noise, they have a limited range of delay times, they use tape and mechanical drive parts which wear, and they are often quite bulky. The music industry had to wait for the by now fast-moving electronics industry to come up with a more convenient storage medium - a 'solid state' medium.

Analogue or Digital?

Sound can be stored digitally in electronic memories too. Random Access Memories (RAMs), just like the ones used in home computers for storing programs, can be filled with digitised sound which, after a delay, can be retrieved and regurgitated in its original analogue form. The snag is the overhead of all the electronics needed to convert from analogue to digital form and back again. In any analogue system, a continuously varying voltage is used to represent the audio signal.

Digital Delay

In a Digital Delay Line (DDL) however, the sound is first divided up into individual time 'packets', as many as 40,000 or so for every second of sound. Each packet is analysed and a binary 'word' of between 8 and 16 bits (Binary digITS) is assembled, whose value is calculated to represent the equivalent analogue value at that time instant. Each of these binary words can then be easily stored in RAMs. This whole process is known as Analogue to Digital Conversion (ADC), and a reciprocal process, Digital to Analogue Conversion (DAC) is needed to restore the binary data into sound. Additional circuitry is needed to control the routing of the data, and to determine which memory location is being written (recorded) into, and read (played) from. Figure 2 shows the schematic for a DDL.

Bucket Brigades

Bucket Brigade Devices (BBDs) are another example of 'solid state' storage. Here though, the signal remains in its analogue form, although it is still divided into time packets.

The device is effectively a chain of small capacitors, each capable of storing the analogue voltage appropriate for each time packet. The capacitors are nested in a network of electronic switches which cause the charges in each capacitor to be shifted in a continuous flow through the chain. This can be likened to a row of firemen passing water along the line by tipping it from bucket to bucket. This system completely eliminates the need for ADCs and DACs or location addressing circuits.

Bandwidth

Common to both DDLs and BBDs is the need for limiting the bandwidth of both the audio input to and output from the delays by low pass filtering to prevent whistles and other unwanted noises from occurring as a result of interaction of the input signal with the 'clock' frequency which is used to divide the signal up into time packets. Unfortunately, long delays are usually achieved by slicing the sound into larger and fewer time packets to get more sound into a given storage capacity. This means that a lower clock frequency is used, so more filtering is needed, resulting in a poorer frequency response. The designer of such a system is therefore usually faced with a compromise between delay length and signal bandwidth.

We still haven't answered the original question; analogue or digital? Well, as you might expect there is no straightforward answer. As far as performance goes, the reader might be misled by the present 'digital mania' into thinking that digital is automatically better. On the contrary, a well designed BBD system could have just as great a dynamic range, just as wide a frequency response, and just as good a noise performance as a budget DDL. Indeed, a BBD system would be completely devoid of the quantisation noise which plagues low cost DDLs. This noise is produced as a result of the process of 'fitting' to every time packet the nearest digital word which, due to the limited resolution of the system, will not be exactly the same as the original analogue value. The error produces quantisation noise in the form of a 'fizzing' sound when a signal is present.

Sound Deterioration

The real problem with BBDs is that of deterioration of the signal when long delay lines are used. The deterioration of noise performance with the longer delay lines can be overcome by using companding techniques, but the loss of signal amplitude, and what is worse 'amplitude stability', with the longer delay lines puts a physical restriction on the maximum line length. For the longer delays then, the choice will necessarily be digital.

For shorter delays, the constraints are not so much physical but economical since the cost overheads of the ADC and DAC precludes the use of a DDL for short delays. Certainly for delays up to 50 milliseconds, the choice would always be analogue, even for professional quality (so long as the 'digital' label does not artificially swing the balance). Budget effects might use BBD delays up to 200 ms or so, but anything longer would almost certainly call for a DDL.

The reader may get the impression from the above that the author is somehow opposed to digital audio. This is definitely not the case; indeed the author has been actively involved in the design of digital audio systems (remember the E&MM Transposer?). What must be stressed though, is that equipment does not have to be digital to be good. A well-designed BBD effect is easily capable of outperforming many budget digital units, so don't be fooled by the jargon!

Okay, so now we have our solid state storage medium; what can we do with it?

Echo & Reverberation

Just like the original magnetic tape machines, echo is achieved by putting the source signal through a fixed delay device, the output of which is mixed back into the input, only to be delayed again and again. The overall gain from the output of the delay device back to its input must be slightly less than unity for the echo effect to decay naturally. If this is not so, then the sound will keep on increasing in level, bringing up all sorts of other noises with it, till a point is reached where the system clips the signal, resulting in a very unpleasant chaos of 'crunchy' sounds.

We have seen numerous echo machines based on analogue (BBD) technology, many of which unfortunately use BBDs which are too short to achieve a reasonable bandwidth at the long delays expected of them. A badly designed analogue echo machine will lack clarity and brightness, and will more than likely exhibit plenty of noise too, so if you are thinking of buying one, have a good listen first to make sure the designer has made the right compromises.

You cannot fail to have noticed the surge of DDLs coming onto the market over the last few months. Most of them seem to offer similar specifications, such as delay times up to a second or so, bandwidth between 8kHz and 16kHz, often switchable for longer delays at the expense of bandwidth. Modulation facilities are usually available too, so that the delay time can be automatically swept to produce flanging and chorus effects... more of this later.

Pseudo reverberation, sometimes known as 'hard reverb' or 'flutter echo' can be produced on most DDLs, although quite effective when used with subtlety at a low volume level in a mix, does sound rather clinical due to the constant echo density, unlike natural reverb where the echo density increases with time.

True digital reverb systems started life, as most audio effects do, in the studio where their magnificent sound quality could be appreciated (and astronomical price tag afforded!). These beasts simulate room reverb by utilising a multiplicity of discrete delays, each with a unique and unrelated delay time. Using complex feedback paths and mixing arrangements (invariably under microprocessor control), a range of convincingly natural reverb effects with highly diffused reflections are achieved. Echo density is the figure of merit for reverbs since sparsely and regularly distributed reflections sound so unnatural. Digital reverbs are starting to fall within the price range of home studio fanatics now, but again watch out for the compromises!

Theoretically, BBD technology should be capable of yielding the delay times needed for reverb (typically under 50ms), and so long as several different unrelated delay times are used simultaneously, then good quality, natural sounding reverb will be obtained which should beat the pants off many spring line systems, and compete strongly against digital reverbs, but no commercial exploitation of this technique 'springs' to mind, if you pardon the pun!

Changing Names

Contrary to popular belief, flanging was not accidentally discovered by a drunken tape op leaning on the moving 'flange' of a tape reel, neither does it derive its name from tape reel flanges. 'Flanger' was in fact the nickname given by the Beatles to the original ADT (Artificial Double Tracking) technique developed in 1964 by Ken Townsend of Abbey Road studios, under the direction of George Martin, the Beatles' producer. This nickname was coined when an inquisitive John Lennon enquired of George Martin the basis of the effect, received the tongue-in-cheek reply 'it's a double bifurcated sploshing flange'!

Flanging as we know it today, was not discovered until 1967 when George Chkiantz calculated that this would be a side effect of a synthetic tape reverberation technique he was working on at the time. He demonstrated the effect in the presence of Glyn Johns, who immediately pressed the effect into service on the Small Faces single, Itchycoo Park. Just to confuse matters further, the effect was originally known as 'phasing'.

The device we now know as a 'phaser' was in fact the earliest form of electronic implementation of the complex studio phasing technique. In the absence of any practical electronic delay line in those early days, phase shift circuits were used in an attempt to simulate the effect. Electronic phasers have thus never sounded quite the same as the original studio phasing (now flanging!).

Flanging

Flanging (or phasing as it was in those days) was originally produced using the setup shown in Figure 4. The two tape machines simultaneously recorded the sound to be processed. The replay outputs of the machines would then be combined so that cancellations would occur at certain frequencies, dependent on the difference in synchronisation between the two outputs, such that 'notches' would occur at regular frequency intervals throughout the audio spectrum.

Since one of the machines would be equipped with varispeed, highly controllable differences in synchronisation up to 20 ms or so could be achieved. The effect of altering the varispeed would be to change the frequency interval between the notches which, when swept up and down would produce the familiar 'sky-riding' sound effect. Although this set-up could produce the effect almost in real time, the delay caused by the record to replay head distance made this impractical for most applications.

Today, flanging is simply a matter of passing the signal through a delay line, usually a BBD, to give up to 20ms of variable delay. Apart from the occasional manual 'sweep' control found on delay units, modulation is invariably performed using a sweep oscillator adjustable both in frequency and sweep depth. The output of the delay line is then mixed with the original signal to produce the characteristic cancellations and notches.

A network configured to give a series of notches is sometimes called a comb filter, and referring to the frequency response of a flanger shown in Figure 5 it is easy to see why.

This real-time solution has the additional advantage that some of the output from the system can be mixed back into the input to provide positive feedback, deepening the notches and making the effect more pronounced. BBDs fit the bill admirably, although the sweep range is sometimes limited in budget units.

Phasers actually employ phase-shifting rather than time-shifting techniques to create phase cancellations and hence notches. Unlike the flanger, which will produce typically between 10 and 100 notches (the number being equal to the product of bandwidth and delay), the phaser only produces typically 3 notches (half the number of phase shifting stages), so the effect is less pronounced. Although the phaser has been looked on by some as a 'poor man's flanger', it produces a sound which has its own unique smooth character, surely earning it a place in today's outboard effects rack.

ADT & Chorus

ADT (Artificial Double Tracking), originally known to the Beatles as flanging (if you have been following the story so far!), was first implemented by Ken Townsend at Abbey Road using two tape machines set up as shown in Figure 6.

A tape of the source music would be loaded onto machine 'A' and played so that the sync output of machine 'A' would be recorded onto machine 'B' whose replay output would be simultaneously mixed with the replay output of machine 'A'. If both machines had the same record (sync) to replay head distances and tape speeds, then both replay signals would arrive at the same time. However, machine 'B' would be equipped with varispeed so that timing differences of up to 40ms or so could be achieved, giving the double image effect characteristic of a 'doubled' track. If you want to hear a stunning example of early ADT, just listen to the vocals on the Beatles' 'All You Need is Love' track.

Again, technology came to the rescue to achieve an easy, real-time effect using a delay line to provide up to 40ms of delay. The output of this, when mixed with the original signal produces exactly the same effect as the original tape set-up. The delay is usually modulated to give the feel of natural randomly varying synchronisation.

Chorus is a very similar effect, except that the delay time is usually around 10ms. The aural effect is very difficult to describe, but the idea is not really to give the impression of a double instrument image by allowing the listener to hear timing differences, but rather to achieve the thick, rich texture of a 'doubled' instrument. Modulation is again used to animate the effect, usually by means of a regularly cycling oscillator with variable frequency.

Doubling, although seldom heard these days, is a form of ADT using delays up to 100ms or so to produce double voicings which are perceived as being obviously synthetic. Again BBDs would be the order of the day for a dedicated effect, but the facility would be available on most DDLs anyway.

Figure 7 shows the general arrangement for a delay line to produce flanging, ADT, chorus or doubling. The delay line could use a BBD or a DDL, although it is generally accepted now that dedicated effects invariably use BBDs, whereas units employing a DDL usually offer a wide range of delay related effects.

Harmonising

This, is where things start to get really interesting, since with a harmoniser we can actually transpose the pitch of an instrument or singer in real time, by an adjustable constant interval, typically up to one octave in either direction. This allows us to probe the exciting areas of automatic real time harmonies and detuned vocals, not to mention Pinky & Perky voices!

In a normal DDL, the digital data representing the analogue input is written into memory and, at the same time, old data is read out in synchronism and converted back to analogue. Both writing and reading are thus controlled by a single clock. In a harmoniser however, reading and writing are controlled by separate clocks. The read clock is adjustable in frequency so that the 'playback speed' is controllable, although the 'record speed' is fixed. You could liken this to playing a tape on a tape machine at a different speed to that at which it was recorded to alter the tuning, except that a harmoniser does it in real time, and without changing the tempo.

It is obvious that if you 'play' more slowly than you 'record', then some of the sounds will be left unplayed. Similarly, when shifting upwards, 'playing' more quickly than 'recording' necessitates playing some parts of the sound more than once to fill out the gaps. This is all taken care of automatically in the harmoniser, but gives rise to 'glitches' where the segments of sound are joined together. Up-market harmonisers use elegant de-glitching circuits to silently 'splice' the segments. De-glitchers are rather complex though, and when purchased as an option, can cost almost as much as the rest of the harmoniser!

The Future

We have started to see the progress in DDLs; more modulation facilities, multiple delays, longer delays, wider bandwidth, wider dynamic range, all for less and less money! A one second delay with quite respectable bandwidth is not uncommon now, and the figure is gradually creeping upwards. There is after all no real physical constraint over the maximum delay achievable with DDL techniques; the barrier is an economic one. RAM is the raw material in question, and it's getting more dense and less expensive every day, mainly due to the ever increasing demand of the computer industry.

Indeed, the memories of home computers are being used as audio storage, using addon converters and controllers such as the E&MM digital signal processor, and other similar ready-built systems. These also make use of the computer's power to 'juggle' and process the sounds for echo, harmonising, sampling with keyboard control etc. It may be that the market for these add-ons will be short-lived as DDLs start to boast more and more facilities, including perhaps sampling with keyboard control.

We may see a new, as yet unknown form of 'solid state' storage one day, perhaps an organic memory, which might offer even more storage, with easier (self) addressing, for even less money. Huge sums of money are being spent on research into new forms of active devices for the computer industry, so it is just a matter of time. Very long delays can at the moment be handled by disk storage, although it would be a nice idea to get away from these mechanical devices. Perhaps bubble memories might take over this role.

The difficult compromise between delay time and bandwidth, although being eased somewhat by cheaper RAM, may soon be eliminated completely using 'switched capacitor filters.' These relatively new analogue devices allow steep cut-off filters to be very simply implemented. The real beauty of them is that their cut-off frequency can be controlled over a wide range by varying a high frequency clock. If this clock is kept at a fixed multiple of the DDL clock, then the bandwidth will always be optimised for a given delay. This feature, I am sure will eventually become standard on all DDLs, and some BBD delays too.

One final thought; it took four years after the invention of ADT to discover the very closely related effect we now know as flanging. What other effects or psychoacoustic phenomena are lurking undiscovered, awaiting the application of the humble delay line?