BPS to Hz Calculator

Bits per second and hertz measure different things

A bit rate counts how many bits move past a point each second. A frequency in hertz counts how many times per second a waveform repeats a cycle. The bit itself is a unit of information, and the hertz is the SI derived unit for one cycle per second; the BIPM SI Brochure defines the hertz as s^-1, and NIST lists the kilo, mega and giga prefixes that turn bps into kbps, Mbps and Gbps. Nothing in those definitions converts one into the other, because the missing piece is how many bits ride on each cycle.

Three numbers describe any digital link:

• Rb, the bit rate in bits per second.
• Rs, the symbol rate in baud, one symbol per baud.
• B, the bandwidth in hertz that the signal occupies.

They connect through k, the number of bits carried by one symbol, with k = log2(M) for M signal levels. The symbol rate is Rs = Rb / k, and for an ideal baseband channel the Nyquist limit gives B = Rs / 2 = Rb / (2k): a channel of width B can pass at most 2B symbols per second. Binary NRZ has k = 1, so its conversions are the easiest to trust: divide the bit rate by 2 and you have the hertz of the fundamental.

Those figures are floors for a mathematically ideal channel. A real link adds filter roll-off, coding overhead and guard bands, so the hertz a radio or cable standard must reserve is wider. The sections below work through each widening step.

Binary NRZ: 1 bps is 0.5 Hz of fundamental

Take the slowest-looking data pattern, an endless string of alternating bits: 1010. Each bit lasts Tb seconds, the waveform repeats every two bits, so the fundamental tone of that square-ish wave sits at one cycle per 2Tb, which is Rb / 2. That tone is what the hertz figure on this page measures.

Worked examples at binary NRZ, k = 1:

• 1,000 bps: 1,000 / 2 = 500 Hz
• 9,600 bps: 9,600 / 2 = 4,800 Hz
• 56,000 bps: 56,000 / 2 = 28,000 Hz

The same rule reads backwards: 1 Hz of ideal baseband bandwidth carries at most 2 bps with binary symbols. A 500 Hz channel tops out at 1,000 bps before any coding, filtering or multi-level trickery. That ceiling is the Nyquist rate, and no modulation scheme defeats it; multi-level signalling changes how many bits each symbol carries, not how many symbols the channel passes.

Keep one caveat in mind: the fundamental is not the whole spectrum. An NRZ waveform spreads energy in a sinc-shaped skirt whose first null sits at the full bit rate, not at half of it. Standards therefore budget filter shapes and margins beyond the Rb / 2 floor, which is exactly what the roll-off factor in section 4 prices in.

Modulation packs bits into symbols: baud = bps / k

Multi-level modulation is the first lever that decouples bits from hertz. If each symbol carries k bits, the symbol rate falls by that factor and so does the bandwidth. The modulation name usually encodes M, the number of levels, and k = log2(M): QPSK has 4 phases and k = 2, 16-QAM has 16 states and k = 4, 256-QAM has 256 states and k = 8.

The table below fixes the bit rate at 10,000 bps and varies only the modulation. The roll-off column uses alpha = 0.35, a common raised-cosine choice.

Check one row by hand. QPSK at 10,000 bps gives 10,000 / 2 = 5,000 baud; Nyquist halves that to 2,500 Hz; widening by the factor 1.35 gives 2,500 x 1.35 = 3,375 Hz. The same 10,000 bps on 256-QAM fits in under 850 Hz, roughly eight times tighter than BPSK.

The price is signal quality. Each doubling of M asks for roughly 3 dB more signal-to-noise ratio to keep the same error rate, so 256-QAM works on a clean cable plant and fails on a noisy wireless hop. When a datasheet quotes bits per second per hertz, the modulation choice is doing that work.

Roll-off widens the band: B = (Rs / 2) x (1 + alpha)

A brick-wall filter is impossible to build, so pulse-shaping filters let the spectrum roll off gradually. The raised-cosine family describes that skirt with alpha, the roll-off factor, which runs from 0 (brick wall, unbuildable) to 1 (doubled bandwidth, tolerant of cheap filters). The occupied bandwidth becomes B = (Rs / 2) x (1 + alpha).

The table holds the symbol rate at 5,000 baud, which is QPSK at 10,000 bps, and sweeps alpha:

Two habits make this formula safe. First, alpha multiplies the symbol rate, never the bit rate: forgetting to halve for Nyquist first is the most common slip. Second, a radio link is passband, not baseband, and a passband carrier needs twice the baseband width, so the RF channel is B_rf = Rs x (1 + alpha), one multiplication more.

Standards pick alpha as an engineering compromise. Satellite and cable systems sit near 0.15 to 0.35 because spectrum is sold by the hertz, while test instruments and some military links use 0.5 or 1.0 to tolerate cheaper filters. When you reverse-engineer a channel width from a bit rate, guessing alpha is where most of the uncertainty lives, and a range of answers is the honest result.

Worked examples from 300 bps to 10 Gbps

The full chain, bit rate to hertz, is three steps: divide by k for the baud rate, halve for Nyquist, multiply by (1 + alpha) for roll-off. Here it is on real systems, with the arithmetic shown.

• 300 bps, binary: 300 / 2 = 150 Hz at the fundamental. No modulation or shaping needed to state the floor.
• 9,600 bps, binary: 9,600 / 2 = 4,800 Hz.
• 1 Mbps, QPSK, alpha 0.25: 1,000,000 / 2 = 500,000 baud; 500,000 / 2 = 250,000 Hz; 250,000 x 1.25 = 312,500 Hz.
• 10 Mbps, 16-QAM, alpha 0.2: 10,000,000 / 4 = 2,500,000 baud; 2,500,000 / 2 = 1,250,000 Hz; 1,250,000 x 1.2 = 1,500,000 Hz, or 1.5 MHz.
• 100 Mbps, 64-QAM, alpha 0.15: 100,000,000 / 6 = 16,666,667 baud; / 2 = 8,333,333 Hz; x 1.15 = 9,583,333 Hz, about 9.6 MHz.

Ethernet standards make the same trade with line codes. 1000BASE-T sends 125 Mbaud on each of four pairs using five-level PAM, and Cat5e cable is rated to 100 MHz. Check the numbers: 125,000,000 / 2 = 62,500,000 Hz of Nyquist floor, and 62,500,000 x 1.6 = 100,000,000 Hz, so the cable rating absorbs the equivalent of a 0.6 roll-off. 10GBASE-T runs 800 Mbaud per pair on Cat6a, whose 500 MHz rating matches 800,000,000 / 2 = 400,000,000 Hz widened by the factor 1.25, an alpha of 0.25.

Notice the pattern across every row: the hertz figure always lands within a factor of about 2 of the bit rate in bits per second, sometimes far below it with dense modulation. Any answer outside that band means a unit slip, almost always bytes mistaken for bits or a decimal prefix misread.

Manchester coding doubles the symbol rate

Line codes change the answer even when the modulation stays binary. Manchester coding, used in classic 10 Mbps Ethernet, forces a transition in the middle of every bit, so each bit is transmitted as two half-bit chips. The symbol rate is therefore Rs = 2 x Rb, and the bandwidth floor doubles with it: B = Rs / 2 = Rb.

The payoff is a self-clocking signal. The guaranteed mid-bit transition gives a receiver a timing edge on every bit, which is why 1980s hardware accepted the doubled bandwidth for reliability.

This is where the famous 10 MHz figure comes from: 10 Mbps Ethernet over Manchester needs a 10 MHz fundamental for the alternating pattern, since every bit carries a transition. The IEEE 802.3 committee knew the cost when it picked the code; later standards swapped line coding for multi-level signalling to squeeze the same bits into narrower cables. The IEEE standards library documents the evolution from 10BASE-T through 10GBASE-T in the 802.3 series.

Practical takeaway: if a problem states a line code, use the symbol rate it implies before touching Nyquist. Manchester doubles, 4B/5B maps 4 data bits into 5 code bits for a 1.25x expansion, and 8b/10b does the same in a 10-bit frame. Forgetting the code and dividing the bit rate by 2 underestimates the bandwidth by exactly the code's expansion factor.

Spectral efficiency: bits per second per hertz

Divide the bit rate by the occupied bandwidth and you get spectral efficiency, the single number that summarizes how hard a link works its spectrum. With the formulas on this page it collapses to Rb / B = 2k / (1 + alpha), so modulation and roll-off are the only variables.

Real systems land below the table because pilots, training symbols and forward error correction consume capacity that carries no payload. A Wi-Fi 6 link on 1024-QAM quotes a phy rate near 1.2 Gbps in a 160 MHz channel, which is under 8 bits/s/Hz after coding, well under the 2k/(1+alpha) ceiling for its modulation.

The hard ceiling above every entry is Shannon's: capacity C = B log2(1 + SNR), with the signal-to-noise ratio inside the logarithm. No modulation choice beats it. A 64-QAM link claiming 10 bits/s/Hz needs an SNR above roughly 27 dB just to satisfy Shannon, which is why dense modulation appears indoors and on cable, and fades out on long radio paths where the noise floor wins.

Use spectral efficiency as a sanity check in the other direction too. If a datasheet claims 40 Mbps in a 200 kHz channel, that is 200 bits/s/Hz, a figure no real link at ordinary signal power can reach; something in the claim, usually the bandwidth number, has been misstated.

Checks that catch the usual unit mistakes

Every wrong bps-to-Hz answer traces back to one of a handful of slips. Run this checklist before trusting a number.

Two habits catch almost everything. First, track the units through every step: bits per second, divided by bits per symbol, gives symbols per second; halved, gives cycles per second. Second, hold the answer against the spectral efficiency table. A binary link at 2 bits/s/Hz, QPSK near 3, 64-QAM near 10: if your claimed pairing falls far outside those, the arithmetic or the assumptions are wrong somewhere upstream.

Finally, remember which direction each question runs. Bits per second to hertz asks "how much spectrum do I need", and the answer is a floor that real hardware exceeds. Hertz to bits per second asks "how many bits fit", and the honest answer is a range that depends on modulation, code rate and noise. This calculator takes the first direction; treat any single-number answer to the second with suspicion.