An analogue signal is a physical imitation of a sound. A digital one is a list of measurements of it. Almost everything that behaves differently between the two follows from that single distinction.
The word is analogue, and it is worth taking literally. When a microphone turns a sound into a voltage, that voltage rises and falls in the same shape as the air pressure did. It is a physical imitation of the sound — a different substance behaving in an analogous way.
A digital signal is not an imitation. It is a description: an ordered list of measurements, each one a number saying what the voltage was at a particular instant. Nothing about it resembles a sound. You could print it out.
Almost everything that behaves differently between the two follows from that one distinction, and once you have it the rest of this course stops being a list of specifications.
An analogue signal is continuous in two senses at once. It has a value at every instant — there are no gaps in time — and that value can be anything at all within its range, with no smallest step. Between any two moments there is another moment, and between any two voltages there is another voltage.
A digital signal is discrete in both senses. It has a value only at fixed instants, and that value must be one of a finite set of numbers. Everything else about digital audio is a consequence of choosing how often, and how finely — which is the whole of the next lesson.
The device that does the conversion is an analogue-to-digital converter, and it is worth knowing what is inside it because the parts explain the specifications.
Coming back out, a digital-to-analogue converter runs the process backwards: it produces a voltage for each number in turn, and then a reconstruction filter turns that stepped output into a single smooth curve.
That last stage is the one people misunderstand, so it is worth stating plainly now and proving in the next lesson: the reconstructed signal is not stepped. For any given set of sample points there is exactly one waveform that passes through all of them and contains no frequencies above the Nyquist limit, and the reconstruction filter's entire job is to produce that waveform. The staircase you have seen drawn in diagrams is the converter's raw output, not the signal that leaves the machine.
The obvious objection to all of this is that measurements have gaps between them, and a real sound does not. Something happened between one sample and the next, and it was not recorded — so how can the description be complete?
The answer, which the next lesson makes precise, is that the signal is not allowed to do anything interesting between the samples. The anti-alias filter has already removed everything above half the sampling rate, and a signal with no high frequencies in it cannot change quickly. It is not merely unlikely to have wandered off between two samples and come back — it is incapable of it, in the same way a car with a speed limit cannot have visited another city between two photographs taken a second apart.
So for a band-limited signal there is exactly one curve that passes through the samples, and reconstructing it is a calculation rather than a guess. That is the whole content of the sampling theorem, and it is why the word "resolution" is misleading when applied to digital audio: within its band, the description is not an approximation of the wave. It is the wave.
Tape ran professional studios for about forty years and did not lose because it sounded bad. Course 8.1 covers what it was and how it worked; this is about what the crossing changed.
Copying a tape means re-recording a physical signal through a record head, a piece of oxide and a playback head, and every one of those adds noise and distortion that cannot be removed. Engineers counted generations for exactly this reason and combined operations into single passes to avoid them.
Copying a file means duplicating numbers. A file copied a million times is bit-for-bit identical to the original, because nothing has been re-recorded — the description was transcribed, not re-performed. The concept of generation loss simply ceased to apply, and an entire discipline built around avoiding it became unnecessary within about a decade.
A tape edit was a razor blade, and the removed material was a strip of plastic on the floor. A digital edit is a change to a list of instructions about which numbers to play in which order, and the numbers themselves are untouched. Undo is not a convenience feature bolted on afterwards; it is a natural property of editing a description rather than an object.
Tape had a sound — a specific list of measurable imperfections that engineers of the era spent real money trying to reduce. A correctly working converter is supposed to have no sound at all: what comes out should be what went in, and any audible contribution is a defect.
Whether that is a gain is a genuine argument rather than a settled question, and it is the reason a large industry exists modelling the imperfections that were removed. But it is a different argument from the three practical ones above, and it should not be mixed up with them.
This is the change that most affects how you actually work, and it gets the least attention.
Analogue levels are referenced to a nominal operating point — 0 VU on a tape machine, which the previous course established is a flux level on the medium. There is usable room above it. Push a tape past 0 VU and it saturates gradually, adding harmonic distortion and rolling off the top, and the whole industry learned to use that region deliberately.
Digital level is measured in dBFS — decibels relative to full scale — and full scale is not a reference point. It is the largest number the format can represent. There is nothing above it, not a small amount of room, nothing. Ask for a larger value and the converter returns the largest one it has, repeatedly, and the result is a flat-topped waveform full of harmonics that were never in the music.
Every analogue habit about pushing a medium for character fails at this boundary, and the transition period is full of records that were made by engineers who had not internalised it yet.
An analogue chain degrades. A slightly dirty connector adds crackle; a failing cable adds hum; a worn tape loses top end. All of it is audible early, gets worse gradually, and can usually be traced by listening while you wiggle things.
A digital link does not do this. It is bit-perfect right up to the point where it is not, and then it drops out, clicks, or delivers nothing at all. There is no region where a digital cable is "a bit worn" and sounds slightly duller — either every bit arrives or some do not, and the difference between those two states is dramatic.
This cuts both ways in practice. It means a cheap digital cable that works is genuinely as good as an expensive one that works, which saves a great deal of money. It also means the diagnostic skill of listening for gradual degradation, which is most of how analogue faults are found, does not transfer at all.
An analogue chain has essentially no delay: the signal propagates at the speed of electricity and arrives when it arrives. A digital chain does not work on a continuous stream — it collects samples into blocks and processes them a block at a time, because that is dramatically more efficient than handling them individually.
Every stage that buffers contributes its block to the total, so the round trip from microphone to headphones has a real, measurable delay that varies with your buffer setting. Small buffers mean less delay and more processor load; large buffers mean the opposite. Nothing in the analogue world required that trade to be made, and every musician who has complained about monitoring latency is complaining about a consequence of blocks.
A sample is two things: a value and an instant. The file stores only the value, because the instants are assumed to be perfectly evenly spaced — the clock's job is to make that assumption true.
If the clock is unsteady, samples are taken slightly early or slightly late, and the values recorded are therefore the values that the signal happened to have at the wrong moments. That error is jitter, and the important part is that it becomes permanent. The file has no record of when each sample was actually taken, so a timing error at the input becomes a value error in the data, and no later process can distinguish it from real signal.
This is why clocking matters at all, and it is a distinct problem from the synchronisation covered in Course 7.3 — that was about machines agreeing where they are in a song, and this is about a converter agreeing with itself about when a microsecond has passed.
People talk about converters as though the quantiser were the whole device, and it is the least interesting part of it. A converter is four things in a box:
Which means most of what anyone describes as "the sound of a converter" is the analogue electronics wrapped around it and the quality of its clock, not the conversion. Two units with identical published specifications can measure and sound different for entirely analogue reasons, and this is worth knowing before spending money on the basis of a bit depth.
A digital copy is exact. A digital path is only exact if nothing along it touches the numbers, and several things commonly do without announcing themselves: an operating system volume control applying gain, a driver silently converting sample rate to match a shared setting, a media player applying its own processing, or a bit-depth reduction on export with no dither applied.
None of these are audible as an obvious fault. They are the difference between a chain that does what you think and one that does something slightly else, and the only way to know is to check what each stage is set to rather than assume.
Once the numbers exist they have to be stored, and it is worth knowing that the common formats are containers rather than encodings.
A WAV file is a header followed by raw PCM samples. The header says what the sample rate is, how many bits per sample, how many channels, and how long the data runs — because the numbers themselves carry none of that. A list of values is meaningless until something states how fast to play them. AIFF is the same idea with a different header layout, and the audio inside both is identical.
Broadcast WAV adds one thing that matters enormously in practice: a timestamp saying when the recording started. That is what allows a field recording to place itself at the correct position on a timeline automatically instead of being lined up by hand, and it is why sound for picture standardised on it.
It is also worth separating two words that get used interchangeably and mean different things:
A lossless file is compressed and loses nothing; an uncompressed file is larger and gains nothing over it. The only real arguments for uncompressed working files are decoding cost and universal compatibility, and both are weaker than they were.
It is worth being blunt, because the converter attracts an amount of attention out of all proportion to its position in the chain.
Digitising changes nothing about the microphone, the room it is in, the position it was placed in, the preamp it went through, or the performance being captured. Those four things account for most of the difference between a recording that works and one that does not, and all four are upstream of the point this lesson is about.
The honest summary is that conversion made distribution, copying, storage and editing enormously better, and made the capture itself marginally better at the top end of the market and dramatically cheaper everywhere else. It did not make anybody's room sound good.
Everything above treats "how often" and "how finely" as settings that exist. The next lesson works out what they actually buy — why the sample rate has to be more than twice the highest frequency and what happens when it is not, why each extra bit is worth about 6 dB, and why adding a small amount of noise on purpose is the correct thing to do rather than a compromise.
The three lessons after that are about what happened when those numbers had to be put onto something you could sell.
Digital does not fail gradually. It is exact until it is not, and then it is a click. Analogue habits about pushing a medium for character do not survive the crossing.
Students pay for getting unstuck: ask a concept question, routing issue, DAW confusion, or mix decision tied to this lesson.
Free readers can learn from the public Q&A archive. Paid students can ask their own lesson-specific questions and get mentor replies.