Frequency Modulation

Understand the enemy before designing around it.

1. Generate an amplitude-modulated carrier and view it on the oscilloscope — a sine wave whose envelope carries the audio.
2. Inject a brief impulse into the signal path, simulating a spark or lightning.
3. Detect the AM and listen to the recovered audio.
4. Now look at the envelope on the scope during the impulse.

The impulse rides directly on the envelope, and because an AM detector recovers exactly that envelope, the click passes straight through to the listener. No amount of filtering after detection helps, because by then the click and the programme are the same kind of thing.

Build the transmitter side and watch the carrier move.

1. Build a stable LC oscillator at a convenient frequency — a few MHz is easy to work with.
2. Include a varactor diode as part of the tuning capacitance — a reverse-biased diode whose junction capacitance varies with applied voltage.
3. Apply an audio signal to the varactor bias and observe the oscillator output on a spectrum analyser.
4. Measure how far the carrier moves for a given audio amplitude.

The carrier amplitude stays constant while its frequency sweeps back and forth in step with the audio. That swing is the deviation, and it — not the audio amplitude — is what now carries loudness.

Deviation is a design choice, and it is the whole argument. Broadcast FM uses ±75 kHz to carry 15 kHz of audio: five times more spectrum than the message strictly needs. That deliberate excess is precisely where the noise advantage comes from, and it is why the narrow-band analysis of the 1920s concluded FM was pointless.

FM spreads energy in a way that surprises everyone the first time they see it.

1. Modulate with a single steady audio tone and view the spectrum.
2. Count the sidebands either side of the carrier and note their spacing.
3. Now increase the deviation while keeping the audio frequency fixed, and watch the pattern change.
4. Find a deviation at which the CARRIER itself disappears.

FM produces a theoretically infinite series of sidebands spaced at the modulating frequency, whose amplitudes follow Bessel functions — and at certain modulation indices the carrier amplitude passes through zero entirely. Seeing the carrier vanish while the signal is still being transmitted is genuinely startling, and it is the standard way to calibrate deviation precisely: adjust until the carrier nulls, and you know the index exactly.

In practice the sidebands beyond a certain point are negligible, and Carson’s rule — bandwidth is roughly twice the sum of deviation and highest audio frequency — captures where to stop. For broadcast FM that gives about 180 kHz, close to the 200 kHz channel spacing actually allocated.

This stage is where the whole advantage is collected.

1. Build a high-gain amplifier stage that clips hard at both extremes — deliberately overdriven, or a pair of back-to-back diodes across the signal.
2. Feed it an FM signal with amplitude noise added.
3. Look at the output on the scope.
4. Measure the residual amplitude variation before and after.

The output is a constant-amplitude square-ish wave; every trace of the amplitude noise has gone, and the frequency information is completely untouched. That is the trick in one sentence — deliberately destroy the dimension the noise lives in, because you are not using it.

An AM receiver cannot do this, because the amplitude IS the message. FM can, because amplitude carries nothing.

Turn frequency variation back into amplitude variation, which is the one thing a detector can handle.

1. Build a tuned circuit and deliberately tune it OFF the carrier frequency, so the carrier sits on the sloping side of its response.
2. Feed the limited FM signal into it and detect the output with a diode.
3. Listen to the recovered audio and measure its distortion.
4. Now build a balanced version — two circuits tuned either side of the carrier, their detected outputs subtracted.

The slope detector works and sounds poor; the balanced discriminator works and sounds good. A single slope is only straight over a short span, so large deviations run into its curvature and distort. Two slopes arranged back to back have their curvatures cancel around the centre, giving a long straight region and a null exactly at the carrier.

Notice the shape of the fix: when a transfer characteristic is nonlinear, using two of them in opposition can cancel the nonlinearity while doubling the wanted response. The same idea gives you the push-pull amplifier, the differential pair and the bridge circuit.