Frequency modulation (FM) encodes information by varying a carrier's instantaneous frequency around its rest value. Unlike AM, the FM spectrum is theoretically infinite — an FM-modulated carrier generates an unbounded set of sidebands at every integer multiple of the modulating frequency on each side of the carrier. In practice almost all of the signal energy sits in a small handful of those sidebands, and the rest is negligible. The math that tells you which sidebands matter — and how much — is the Bessel function of the first kind Jn(β).
Modulation index β
For a single-tone modulating signal, the FM modulation index is defined as β = Δf / fm, where Δf is the peak frequency deviation and fm is the modulating frequency. β is dimensionless. A small β (< 0.2) means narrowband FM (NBFM) — the spectrum closely resembles AM with just one significant sideband pair. A large β (> 1) means wideband FM (WBFM) — energy spreads over many sideband pairs.
Carson's rule bandwidth
Carson's rule is the engineering shortcut for FM bandwidth: BW ≈ 2 (Δf + fm) = 2 fm (β + 1). It approximates the bandwidth that contains roughly 98% of the FM signal's power for sinusoidal modulation. For broadcast FM (Δf = 75 kHz, max fm = 15 kHz), Carson gives BW = 2 (75 + 15) = 180 kHz — closely matching the 200 kHz channel spacing actually used in FM broadcasting (with guard bands).
Bessel sideband amplitudes
For a single-tone FM signal s(t) = A cos(2π fc t + β sin(2π fm t)), the Fourier decomposition gives A · Jn(β) for each sideband at fc + n·fm, where n is any integer (positive and negative). Jn(β) is the Bessel function of the first kind of order n. The sideband at the carrier itself (n = 0) has amplitude J0(β), which actually decreases as β grows — at β ≈ 2.405 the carrier vanishes entirely (the first zero of J0), and all energy is in the sidebands.
Total power is constant
A key property of FM: the total transmitted power is independent of β. Mathematically, J0²(β) + 2 · Σ Jn²(β) = 1 for any β. Modulating an FM signal doesn't change its average power — it just redistributes that fixed power across more or fewer sidebands. This is why FM transmitters drive constant-envelope class-C amplifiers, getting ~70% efficiency where AM is stuck at ~50%.
Significant sidebands & the 98 % rule
The traditional cutoff: a sideband is "significant" if |Jn(β)| ≥ 0.01 (1% of unmodulated carrier amplitude). Counting these sidebands and multiplying by 2fm gives a more accurate FM bandwidth than Carson's rule. Carson's rule typically holds 98% of the power but slightly under-counts the high-β extremes; using a stricter 0.1% threshold (regulatory FCC mask, for example) gives a wider bandwidth. This calculator lets you toggle the threshold to see the trade-off.
What's the difference between modulation index β and frequency deviation Δf?
Δf is the maximum amount (in Hz) that the carrier frequency swings away from its rest value when modulated. It's set by the transmitter's deviation circuitry and is regulated (e.g., FM broadcast is legally limited to Δf ≤ 75 kHz in most jurisdictions). β = Δf / fm is dimensionless and tells you how the spectrum will distribute across sidebands. A 5 kHz tone with Δf = 75 kHz gives β = 15 (very wideband); the same Δf = 75 kHz with a 75 Hz tone gives β = 1000 (extremely wideband). FM broadcast uses pre-emphasis to keep the worst-case β manageable.
Why does the carrier disappear at β ≈ 2.405?
2.405 is the first zero of the Bessel function J0(β). The carrier amplitude in an FM signal is exactly A · J0(β), so when β hits a zero of J0, the carrier vanishes and 100% of the transmitted power moves into sidebands. The subsequent zeros are at β ≈ 5.520, 8.654, 11.79, etc. This is actually used as a calibration trick: feed a known tone, sweep deviation up until the carrier on a spectrum analyzer goes to zero — you've just calibrated your transmitter's deviation to 2.405 × fm.
When is Carson's rule wrong?
Carson's rule is a 98%-power approximation. It tends to under-estimate the true bandwidth at very low β (NBFM) where the formula 2(Δf + fm) ≈ 2 fm ignores that one extra sideband pair, and slightly over-estimates for moderate β. For β < 0.2 use 2 fm instead (the actual NBFM bandwidth). For very large β it's quite accurate (the spectrum becomes nearly rectangular over ±Δf). For regulatory work (FCC emission mask), use a stricter cutoff like 0.1% or compute the bandwidth from the Bessel sidebands directly — this tool's "98% Bessel bandwidth" line gives exactly that.
Why does FM use class-C amplifiers but AM does not?
Because FM has a constant envelope — the amplitude of the RF signal never changes, only its frequency. A class-C amplifier is highly non-linear (~70–80% efficient) but only preserves frequency, not amplitude shape. That destroys AM but is perfect for FM. AM transmits its information in the envelope shape, so it needs a linear class-A or class-AB amplifier (~30–50% efficient). The constant envelope of FM also gives it the famous "capture effect" — a stronger signal completely suppresses a weaker co-channel signal at the receiver, which is why FM sounds cleaner than AM in noisy conditions.
How is FM bandwidth in stereo broadcasting different?
Stereo FM uses a more complex baseband structure (L+R from 0–15 kHz, a 19 kHz pilot tone, L−R DSB-SC around a 38 kHz subcarrier from 23–53 kHz, RDS at 57 kHz, sometimes SCA at 67 or 92 kHz). The effective maximum modulating frequency fm,max goes from 15 kHz (mono) up to ~67–92 kHz (full stereo + auxiliaries). With Δf still capped at 75 kHz, β for stereo drops from 5 (mono) to roughly 1 (full multiplex). Carson's rule gives a bandwidth around 2 (75 + 67) = 284 kHz for stereo+RDS, slightly above the 200 kHz channel slot — which is why the modulator must pre-emphasize and shape carefully.
How does this calculator compute the Bessel Jn(β) values?
It uses Miller's backward-recurrence algorithm, which is the standard numerically-stable method for Jn(x) of integer order n. The recurrence Jn−1(x) = (2n / x) Jn(x) − Jn+1(x) is run from a high starting order N down to 0, with the resulting values normalized using the identity J0(x) + 2 Σ J2k(x) = 1. This avoids the numerical instability of forward recurrence (which is well-conditioned only for n ≤ x but blows up for n > x). All computation runs in your browser; nothing is sent to a server.
What's the relationship between FM and PM (phase modulation)?
FM and PM are mathematical duals. For a sinusoidal modulating signal, the spectra are identical — both have Bessel Jn(β) sidebands. The difference is what's proportional to what: in FM, frequency deviation Δf is proportional to the modulating amplitude (so β = Δf / fm depends on fm). In PM, phase deviation is constant with fm, so β doesn't depend on fm. In practice, digital systems (4FSK, GFSK, MSK, GMSK used in Bluetooth, GSM, etc.) blur the distinction by directly modulating an oscillator with shaped pulses; the result is mathematically PM but is often analyzed using FM techniques.