Also known as: roll-off factor, excess bandwidth factor, alpha, beta
Roll-off factor (usually α, sometimes β) is the parameter, between 0 and 1, that sets how gradually a raised-cosine or root-raised-cosine pulse’s spectrum tapers to zero — and therefore how much excess bandwidth the shaped signal uses beyond the theoretical Nyquist minimum.1 It is the main knob a system designer turns when balancing spectral efficiency against how forgiving the link is.
How it works
The narrowest possible zero-ISI pulse is a brick-wall filter occupying exactly the Nyquist bandwidth R_s/2 either side of center (α = 0), but that ideal has an infinitely long sinc-shaped tail and cannot be built. The raised-cosine family fixes this by rounding the band edge with a cosine skirt whose width is set by α: the occupied one-sided bandwidth is B = (1 + α)·R_s/2, so α is literally the fractional excess bandwidth. At α = 0 the spectrum is the unrealizable brick wall; at α = 1 the signal occupies twice the Nyquist bandwidth but has a smooth, easily realized roll-off and short, well-behaved pulse tails.
The choice is a genuine trade-off. Low α packs the signal into less spectrum and improves spectral efficiency, but the pulse tails grow longer and larger, which makes the eye more sensitive to sampling-timing error and raises the signal’s peak-to-average ratio (harder on the transmit amplifier). High α wastes spectrum but yields short tails, a wide-open eye, and relaxed timing tolerance. Because the shaping is split as a root-raised-cosine at both ends, the transmitter and receiver must use the same α for the composite to remain a proper Nyquist pulse.
There is a matching story in the time domain. A low-α pulse is a long, slowly decaying sinc-like waveform with prominent ringing before and after its peak; a high-α pulse is compact, dying away within a couple of symbol periods. Those tails are exactly the energy that becomes intersymbol interference if the receiver samples even slightly off the ideal instant, which is why low-α systems demand tight clock recovery. The long tails also make the transmitted envelope swing more, raising the peak-to-average power ratio and forcing the power amplifier to back off from saturation — a real cost in battery-powered and satellite links where amplifier efficiency matters.
Relevance to SDR
Every digitally modulated signal an SDR touches has a defined roll-off, and matching it is part of building the receive matched filter. GopherTrunk’s demodulators implement an RRC filter whose α matches each protocol: P25 C4FM/CQPSK uses a nominal α ≈ 0.2 root-raised-cosine shaping, and other four-level and PSK trunking modes specify their own values. Using the wrong α mismatches the transmit and receive filters, so the composite is no longer ISI-free and the eye partly closes even on a clean signal — a subtle way to lose sensitivity. Knowing the roll-off also lets a receiver predict a channel’s occupied bandwidth and set its channelizer width accordingly.
In practice
Common values cluster low for spectrum-tight land-mobile systems (α around 0.2) and higher in systems that prize simple filtering (older modems used α = 0.35 or more). Halving α saves only a fraction of bandwidth but noticeably tightens the timing budget, which is why very small α is reserved for links that can afford precise clock recovery.
A related subtlety is filter length. A root-raised-cosine filter is theoretically infinite, so an implementation truncates it to a finite span (commonly ±4 to ±8 symbols) and applies a window. A small α needs a longer truncation to capture its slowly decaying tails without spectral artifacts, so the bandwidth saving of low α is partly paid back in a longer, more expensive filter. GopherTrunk builds its RRC matched filters at a fixed number of taps per symbol chosen to hold the truncation error well below the noise floor for each protocol’s α, so the theoretical roll-off and the realized one agree closely.