skip to content

A bandwidth-starved downlink buys less and less rate for each extra watt of transmit power - why?

level: seniorimportance: should knowfreq 42%

answer

  1. two levers with different curves
  2. bandwidth multiplies, power is logged
  3. one extra bit per power doubling
  4. wider band admits more noise too
  5. power-limited links saturate near 1.44 S/N0

basics

~20 s

Capacity grows with the logarithm of the signal-to-noise power ratio, so each doubling of power adds only about one more bit per second per hertz. Bandwidth sits outside the logarithm as a multiplier, which is why a starved link is starved.

solid answer

~40 s

The Shannon-Hartley form is `C = B x log2(1 + S/N)`: bandwidth `B` multiplies, while the signal-to-noise power ratio `S/N` enters through a logarithm. Doubling power moves `S/N` from 15 to 31 and `log2(1 + S/N)` from 4 bits per hertz to 5 - a 25% gain for twice the power, and the next bit costs another doubling. Widening the band is the linear lever by comparison, but it is not free either: at a fixed noise density the admitted noise power grows with the band, so capacity saturates rather than rising forever. A power-limited, band-limited link is therefore the regime where neither knob is generous, and lowering the code rate to fit under the real ceiling is often the cheaper move.

code

pseudocode · 13 lines
pseudocode
B = 2000000                       # bandwidth in hertz

for each snr in [15, 31, 63, 127]:        # each entry roughly doubles the power
    bits_per_hz = log2(1 + snr)           # 4, then 5, then 6, then 7
    C = B * bits_per_hz                   # bits per second
    report snr, bits_per_hz, C

# holding transmit power fixed while widening the band admits more noise:
S = 1.0
N0 = 0.0000001                    # noise power per hertz
for each width in [B, 2*B, 4*B, 1000*B]:
    C = width * log2(1 + S / (N0 * width))
    report width, C               # rises, then flattens towards S / (N0 * ln 2)

go deeper

for a junior

Remember the two ingredients of a link's ceiling - how wide the band is and how strong the signal is against the noise - and that the second one pays off far more slowly.

for a middle

Explain the formula's shape: bandwidth multiplies while the signal-to-noise ratio is inside a logarithm, so a power doubling is worth about one extra bit per hertz.

for a senior

Diagnose which regime a link is in - band-limited or power-limited - and pick the lever that matches, including the receiver-side and code-rate moves rather than only transmit power.

for a principal

Own the economics: spectrum, energy and hardware quality buy rate on different curves, and past the power-limited knee the right answer may be to redesign the mission rather than the link.

## The shape of the formula For a band-limited link with additive noise, capacity takes the Shannon-Hartley form: - `C = B x log2(1 + S/N)` bits per second, where `B` is bandwidth in hertz and `S/N` is the received signal-to-noise **power** ratio. Everything about the diminishing returns is in the asymmetry of that expression. Bandwidth is a plain multiplier; the power ratio is inside a logarithm. Two levers, two completely different return curves. ## What a doubling of power actually buys Work it at the operating points where the arithmetic is clean: | `S/N` (power ratio) | `log2(1 + S/N)` | Rate at `B` = 2 MHz | |---|---|---| | 15 | 4 bits per hertz | 8 Mbit/s | | 31 | 5 bits per hertz | 10 Mbit/s | | 63 | 6 bits per hertz | 12 Mbit/s | | 127 | 7 bits per hertz | 14 Mbit/s | Each row roughly doubles the power of the row above it and adds **one** bit per hertz. The first doubling buys 25%, the next 20%, the next 17%. On a link that already has a healthy ratio, throwing power at the problem is the expensive lever, and on a remote station with a fixed energy budget it is often not a lever at all. ## Why widening the band is not a free lunch either Bandwidth multiplies the whole expression, so at a fixed `S/N` doubling the band doubles the rate. But holding `S/N` fixed while widening the band is a sleight of hand: noise enters over the whole band, so at a fixed noise power density `N0` the admitted noise power is `N = N0 x B`. Substituting gives - `C = B x log2(1 + S / (N0 x B))` which **increases** with `B` but flattens out: as the band grows without limit, capacity approaches `S / (N0 x ln 2)`, about `1.44 x S/N0`. That saturation point is set by received signal power and noise density alone. A link in that regime is **power-limited**: no amount of extra spectrum helps, because each new hertz brings its own noise. The two regimes are worth naming: - **Band-limited** - plenty of power, not enough spectrum. `S/N` is high, each extra bit per hertz costs a power doubling, and extra bandwidth pays linearly. - **Power-limited** - plenty of spectrum, not enough received power. Capacity is near its `1.44 x S/N0` asymptote and widening further buys almost nothing. ## The knobs, honestly ranked 1. **More bandwidth** - linear while you are band-limited, then saturating. Often the scarcest resource, and shared with everyone else nearby. 2. **More received signal power** - logarithmic. Includes better antennas and shorter paths, not only transmit watts. 3. **Less noise** - the same logarithm from the other side; a quieter receiver front end raises `S/N` exactly as more power does. 4. **A lower code rate** - does not move the ceiling at all, but moves your operating point under it. This is the knob that turns errors into working links when the ceiling itself is fixed. That fourth entry is the one candidates forget. Coding cannot raise `C` - it decides how much of `C` you attempt to use. When a downlink is failing and neither power nor spectrum is available, backing the rate off is the remaining move, and it costs throughput rather than energy. ## What this rules out - **Any claim of unlimited rate over a fixed band.** Rate per hertz grows only logarithmically in power, so "we will just turn it up" has a hard economic wall. - **Any claim that a clever code beats the formula.** `C` bounds every code over the channel; a code that appeared to exceed it would have to be exploiting structure the noise model got wrong. - **Any claim that a wider band always helps.** Past the power-limited knee, extra spectrum contributes noise as fast as it contributes opportunity. The formula also explains a habit that looks strange from outside: engineers quote link margins in decibels because power enters logarithmically, so a decibel is the natural unit of the thing that actually moves the rate linearly. Roughly, in the high-ratio regime, every 3 dB of extra margin is worth about one more bit per second per hertz.

  • If more power barely helps and no spectrum is available, what is left?
    Lower the code rate so the operating point sits further under the unchanged ceiling, accept less payload throughput, and look at the receiver: a quieter front end or a better antenna raises the signal-to-noise ratio exactly as extra transmit power would, often more cheaply on a remote station.
  • Why does capacity not grow without limit as the band is widened?
    Because noise arrives across the whole band. At a fixed noise density the admitted noise power grows in proportion to the bandwidth, so the ratio inside the logarithm falls as fast as the multiplier outside it rises. Capacity approaches about 1.44 times signal power over noise density and stops.

saying these in an interview costs you the question

  • Expects capacity to scale linearly with transmit power
  • Believes a wider band always raises capacity without limit
  • Thinks a better code can push the rate above the ceiling
  • Ignores receiver noise as a lever equal to transmit power
  • Confuses the power ratio with a ratio of amplitudes