| Episodes | 28 |
| Mentions | 33 |
| Cited here | 23 |
| First — last | #6 — #729 |
| Top guests | Greg Charvat, Hank Zumbahlen, Dafydd Roche |
| Related | analog to digital converter · oscilloscope |
Signal-to-noise ratio (SNR) expresses how strongly a desired signal stands above the noise accompanying it, determining whether information can be converted, communicated, recovered, or resolved.[185][352] It functions as an engineering budget: available ratio can be exchanged for data rate, integration time, measurement resolution, modulation complexity, or operating range.[179][427][483][533] Because both the signal and noise sides can be engineered, improving SNR may involve increasing signal energy, reducing noise, narrowing bandwidth, integrating longer, or using a waveform that separates the wanted signal from interference.[107][376][443][560]
Data conversion and measurement
Resolution and accuracy are distinct converter properties: resolution describes how many output bits change, while accuracy describes how many of those bits contain meaningful information.[185] A converter specified as 24-bit therefore has to be interpreted through its signal-to-noise performance; grounding its inputs does not guarantee that its lower bits will remain stationary.[185]
Effective number of bits and SNR contain equivalent information expressed in different units, with the conversion represented by 6.02 times the bit count minus 1.76.[185] The choice between stating performance as effective bits or as a ratio often reflects an engineer’s background rather than a difference in the underlying measurement.[185]
Converters with the same architecture and nominal bit count are not necessarily interchangeable.[65] An audio-oriented converter is optimized for dynamic performance such as distortion and signal-to-noise ratio across its band, whereas a strain-gauge bridge requires accurate DC measurement; using the audio part for the bridge can produce invalid results despite an identical 24-bit delta-sigma designation.[65]
The sigma-delta architecture obtains very high SNR at the cost of latency.[474] This exchange is compatible with audio because delays on the order of a hundred milliseconds are not perceptually important, but the same delay can be unacceptable in a control loop.[474]
Process-voltage scaling directly constrains analog SNR: reduced voltage swing across a resistor produces less signal for the same noise, lowering the achievable ratio.[579] This constraint makes high-performance analog circuitry difficult and expensive to implement on processes optimized primarily for digital devices.[579]
Similar considerations apply inside memory circuits, where a sense amplifier consists of two cross-coupled inverters.[297] Much of the subsequent engineering reduces the amplifier’s offset, improving SNR and thereby increasing the yield of the complete device.[297]
Communications capacity and link design
Shannon’s channel-capacity theorem sets a theoretical ceiling on the number of bits per second that can pass through a channel of specified bandwidth at a specified SNR.[352] The ratio therefore operates as a currency that can be spent on greater throughput.[352]
Because the ratio buys throughput, anything that caps it removes options rather than merely degrading performance. High crosstalk in a connector limits the achievable SNR, and in doing so eliminates signalling strategies that would otherwise have been available — which makes improving the connector, rather than devising cleverer modulation, the change that unlocks a higher data rate.[77]
A link budget is the difference, in decibels, between transmitted power and the receiver’s minimum usable sensitivity.[443] For example, a transmitter supplying 20 dBm and a receiver with a real-world sensitivity of −121 dBm provide approximately 141 dB for distance, antenna effects, and obstructions.[443]
Received signal strength and SNR are separate measurements.[443] A strong wireless signal is generally better than about −60 dBm, but its SNR can still deteriorate when other devices raise the noise level in the same band; the lower interference typical of sub-gigahertz spectrum consequently gives those bands better outdoor behavior than the more heavily used 2.4 GHz band.[443]
Higher-order modulation places more information into each hertz but requires a higher SNR for successful decoding.[430] If that ratio is unavailable, the transmitter and receiver must be brought closer together or the system must spend more resources on processing; exhaustion of these options contributed to the use of millimetre-wave frequencies and beam-forming.[430]
Adaptive radio links continuously trade throughput for margin by selecting lower modulation rates when vehicles, rain cells, or other obstructions reduce path quality.[533] Lower-rate modulation requires less SNR, while the highest-rate modes may operate with the signal only slightly above the noise, causing degradation to appear as intermittent pixelation rather than as an abrupt loss of service.[533]
The same exchange appears at extremely low data rates: a one-megabit-per-second link requires a comparatively good SNR, while a text-message-sized packet sent once per day requires very little.[427] When an application tolerates that low rate, reducing throughput can extend communication into locations inaccessible to a faster link.[427]
A broadband noise source acts as a jammer regardless of intent.[391] Moving the receiver next to the transmitter does not necessarily restore the link, because successful decoding depends on the ratio between signal and noise rather than on absolute received level alone; a coded digital signal cannot be recovered once the noise floor has been raised above it.[391]
Recovering signals from noise
Averaging improves SNR because repeated signal components accumulate in proportion to the number of measurements, while uncorrelated noise accumulates only in proportion to the square root of that number.[483] Sufficient repetitions can therefore reveal a signal that was initially buried below the noise.[483]
Weak-signal radio operation applies the same principle by using minimum bandwidth and modes slow enough that transmitting a call sign can take a minute.[107] Receiver integration over that interval substantially improves noise performance without increasing transmitter power.[107]
Satellite navigation uses processing gain to recover signals below the receiver’s noise floor: each information bit is spread across 1023 chips, and the receiver correlates the received waveform with the known chip pattern.[352] This spreading makes the low-power signal from a distant satellite recoverable despite an initially unfavorable ratio.[352]
In dense urban environments, one transmission can arrive repeatedly after reflections from buildings, producing copies with different delays and amplitudes.[376] Chirped spread-spectrum modulation allows the intended signal to be separated from these reflections, improving the effective SNR without additional transmit power.[376]
A lock-in amplifier exploits advance knowledge of a signal’s frequency to recover a carrier from noise thousands of times larger.[455] It can extract microvolt-level signals from volt-level noise and, in extreme cases, recover signals extending into the nanovolt range.[455]
Reducing each observation to one bit—recording only whether a clock sampled a one or a zero—substantially worsens the SNR of each individual sample.[693] A sufficiently large number of such samples can nevertheless recover information that no individual sample preserves, illustrating the same quantity-versus-quality exchange used by other one-bit receivers.[693]
Large, properly spaced phased arrays provide a direct means of imaging at frequencies where many channels are needed to pull genuine signals out of noise.[729] This physical recovery is distinct from allowing an algorithm to generate a plausible output where no corresponding signal was actually recovered.[729]
Radar and imaging
Through-wall radar can use long, low-peak-power frequency-modulated continuous-wave chirps instead of extremely short, high-power impulses.[115] The principal difficulty is rejecting the very strong reflection from the wall, because that return can saturate the receiver and distort signals from objects behind it.[115]
The impulse alternative requires approximately 10 kW in a two-nanosecond pulse and digitization at several gigasamples per second.[115] Faster digitizers provide poorer performance than slower ones, and the resulting data volume makes real-time processing impractical; long waveforms instead shift more of the problem into signal processing, where it is less expensive.[115]
Angular resolution can also be purchased with SNR.[179] By comparing amplitude and phase across four sub-arrays, a system with a 20 dB SNR can subdivide a nominal ten-degree beam by a factor of one hundred, resolving angles to approximately one-tenth of a degree although the physical antenna alone cannot resolve better than ten degrees.[179]
For a distant target, spot size is proportional to wavelength multiplied by distance and divided by antenna radius, so resolution improves with aperture.[483] When a sufficiently large antenna cannot be carried on an airborne platform, trajectory and integration time become the remaining design variables; flying closer improves resolution but increases relative velocity, shortening the available integration time at the point where more SNR is needed.[483]
Audio and sensing media
Analog playback can contain harmonic distortion that should not be confused with improved fidelity.[270] Measured SNR places vinyl above cassette but below compact disc.[270]
As reproduction equipment has approached the point at which further improvements to the signal produce little practical benefit, the noise side of the ratio becomes more significant.[560] In listening systems, this shifts the remaining engineering problem toward room and environmental noise rather than toward the playback equipment alone.[560]
Image-sensor age and commercial category do not by themselves determine measurement quality.[325] An older sensor technology can retain better noise performance than newer commercial alternatives, making it the more suitable scientific or space instrument where SNR is the governing requirement.[325]