A 200 kHz signal and a channel with roughly 1 MHz coherence bandwidth answer different questions. The first is a property of the transmitted or received signal definition. The second summarizes frequency correlation in the propagation channel for an environment, link, time and threshold.
The practical reasoning is a ratio and a boundary check. When a signal is narrow compared with the channel’s coherence bandwidth, its spectral components may experience similar channel response. When it spans several coherence intervals, different parts can fade differently and a frequency-selective model may be needed.
NTIA’s Institute for Telecommunication Sciences identifies coherence or correlation bandwidth, RMS delay spread and channel impulse response as measured propagation quantities. Its best-practices guide also shows that sounder architecture, measurement bandwidth, resolution and environment shape what can be estimated.
The central fictional channel-sounding result defines coherence bandwidth at the frequency separation where normalized correlation first falls to 0.5. Correlation is 0.76 at 0.5 MHz and 0.43 at 1.0 MHz; linear interpolation for teaching gives about 0.89 MHz. A 0.20 MHz signal is small relative to that value, while a 2.40 MHz signal spans about 2.7 coherence-bandwidth units. The interpretation remains conditional on the threshold, measurement and stationarity assumptions.
This specialist guide belongs to the Wireless Communication root. It also connects to The Gold Standard of Education | Learning That Transfers into Life and the How X Works hub. The worked cases are original teaching examples; their numbers are stipulated so that the reasoning can be checked.
eduKateSG · Wireless Communication
Find your next learning step
Signal bandwidth describes the transmitted waveform or occupied signal under a stated definition. Coherence bandwidth describes how rapidly the propagation channel changes with frequency under a stated correlation rule. Comparing them helps predict flat or frequency-selective behaviour; they are not two measurements of the same object.
ROUTE 1 · CHAPTERS 1–4
Define the two bandwidths
Name signal, channel, threshold and measurement window.
ROUTE 2 · CHAPTERS 5–8
Work the correlation example
Estimate a coherence bandwidth and compare two signals.
ROUTE 3 · CHAPTERS 9–11
Audit the measurement
Connect impulse response, delay spread and sounder limits.
ROUTE 4 · CHAPTERS 12–15
Transfer the model
Test changed thresholds, environments and multicarrier signals.
ROUTE 5 · CHAPTERS 16–32
Report and continue
Repair claims, quantify uncertainty and follow verified sources.
Full chapter contents · Wireless Communication subject guide
Chapter contents
Define the two bandwidths · Chapters 1–4
1. Name the object before the unit
2. Define coherence through frequency correlation
Work the correlation example · Chapters 5–8
5. Inspect the measured correlation table
Audit the measurement · Chapters 9–11
9. Connect frequency variation to multipath delay
Transfer the model · Chapters 12–15
12. Change the correlation threshold
Report and continue · Chapters 16–32
17. Answer focused reader questions
18. Sources, sibling routes and a final transfer calculation
19. Match frequency step to the threshold decision
20. Put uncertainty around the crossing
21. Report distributions across locations
22. Audit the flat-fading model before using it
23. Separate measurement bandwidth from signal bandwidth
24. Practise a boundary-sensitive transfer
25. Keep coherence bandwidth separate from the receiver filter
26. Check correlation magnitude and complex correlation
27. Avoid treating one deep notch as the whole channel
28. Create a reproducible channel-comparison record
29. State what would falsify the chosen channel model
30. Relate coherence bandwidth to equalizer evidence carefully
Both quantities use hertz, but a shared unit does not make them interchangeable. Signal bandwidth describes a spectral extent under an occupied-bandwidth, necessary-bandwidth, null-to-null, emission-mask or other stated convention. Coherence bandwidth describes similarity of channel response between frequencies under a correlation or variation criterion.
Write the comparison as Bsig versus Bc, then annotate both. Bsig needs waveform, filter, allocation and measurement convention. Bc needs link endpoints, environment, time window, motion state, polarization, frequency band, correlation definition and threshold. Without these labels, a neat ratio can compare unrelated summaries.
The reader question is usually predictive: will the channel look approximately flat across this signal, or can different spectral components experience meaningfully different gain and phase? That prediction is conditional, not a permanent label attached to the radio or room.
CHAPTER 2 OF 32 · Define the two bandwidths
2. Define coherence through frequency correlation
Let H(f) represent the complex channel frequency response for a specified link state. A frequency-correlation function compares H at frequencies separated by Δf. Coherence bandwidth is then defined from the range of Δf over which the response remains sufficiently correlated under a chosen threshold.
Different texts and systems use different thresholds or related criteria, such as correlation 0.5 or 0.9. The stricter threshold usually produces a smaller coherence bandwidth. Report the threshold with the number; “Bc=1 MHz” without it may not be comparable with another study.
Correlation is statistical and measurement-dependent. It can be averaged over positions, time or realizations, and the averaging set changes the estimand. A channel that is locally smooth at one position can have a different ensemble correlation across a moving route.
CHAPTER 3 OF 32 · Define the two bandwidths
3. Define signal bandwidth with a named rule
A modulated signal does not always have a sharp spectral edge. Occupied bandwidth can enclose a stated percentage of power; necessary bandwidth relates to emission designation; a filter passband or allocated channel width answers yet another question. Choose the definition that matches the model or requirement.
For the teaching cases, Bsig means the frequency span over which the receiver’s useful baseband waveform carries material energy after the named transmit filtering. It is stipulated as 0.20 MHz for Signal A and 2.40 MHz for Signal B. These are not claimed occupied-bandwidth measurements.
Do not substitute configured channel width for actual signal bandwidth. The existing Wireless Communication specialist branch on channel width versus delivered throughput shows that configuration, mode, overhead and link conditions remain separate. Here the signal span is compared specifically with propagation correlation.
CHAPTER 4 OF 32 · Define the two bandwidths
4. Use the ratio as a model prompt, not a law
The dimensionless ratio β=Bsig/Bc is a useful first prompt. When β is much less than one, the channel response may be approximated as similar across the signal. When β is greater than one, the waveform spans frequency separations that can decorrelate, so frequency-selective effects become plausible.
No universal β cutoff makes every system flat or selective. The acceptable variation depends on modulation, equalization, coding, performance target and the actual channel response. A threshold chosen for one analysis should not be presented as a natural constant.
Retain the underlying response or correlation evidence. Two channels can have the same threshold crossing but different shapes, notches and phase behaviour. The single Bc value is a summary for a stated decision, not a lossless description of the channel.
| Quantity | Fictional definition | Value |
|---|---|---|
| Coherence bandwidth | First separation where correlation falls to 0.5 | about 0.89 MHz |
| Signal A bandwidth | Useful filtered waveform span | 0.20 MHz |
| Signal B bandwidth | Useful filtered waveform span | 2.40 MHz |
| Comparison ratios | Bsig/Bc | about 0.22 and 2.70 |
CHAPTER 5 OF 32 · Work the correlation example
5. Inspect the measured correlation table
Suppose a stationary channel-sounding record produces normalized magnitude-correlation estimates of 1.00 at Δf=0, 0.92 at 0.20 MHz, 0.76 at 0.50 MHz, 0.43 at 1.00 MHz and 0.12 at 2.00 MHz. Sampling uncertainty and processing choices are temporarily omitted to focus on the logic.
With a 0.5 threshold, the crossing lies between 0.50 and 1.00 MHz. The table does not justify pretending that the crossing occurred exactly at either measured separation. A bracket [0.50,1.00] MHz is already an honest estimate; interpolation adds a model assumption.
Plot the points and preserve the complex-response processing that produced them. A monotone-looking table may hide local structure at unmeasured separations. If the decision is sensitive near the threshold, collect or compute finer separations rather than relying on a smooth line drawn for presentation.
CHAPTER 6 OF 32 · Work the correlation example
6. Interpolate the teaching crossing
For a simple teaching estimate, assume correlation changes linearly between (0.50 MHz,0.76) and (1.00 MHz,0.43). The drop needed to reach 0.50 is 0.26 out of the segment’s total drop 0.33. The fraction is 0.26/0.33≈0.788.
Move that fraction across the 0.50 MHz separation interval: 0.50+0.788×0.50≈0.894 MHz. Report about 0.89 MHz, along with the threshold and linear-interpolation assumption. The un-interpolated bracket remains useful evidence about precision.
This arithmetic does not define coherence bandwidth universally. A fitted correlation model, a direct threshold-search procedure or a confidence interval can be more suitable in real studies. The worked value exists to show how the stated definition leads to a reproducible number.
Signal A has Bsig=0.20 MHz. Using Bc≈0.89 MHz gives β≈0.22. Its spectral components are separated by less than the teaching coherence scale, so an approximately flat channel model may be a reasonable starting hypothesis for this record.
A starting hypothesis still needs verification. Inspect H(f) across the actual signal band, including phase, and check the performance metric that matters. A deep notch or rapid phase change inside the band can defeat a simplistic conclusion even when the summary ratio looks small.
State the narrow conclusion: “For the specified measurement, threshold and Signal A span, the channel response is expected to be relatively correlated across the signal.” Do not upgrade it to “Signal A never experiences frequency selectivity” across other locations, times or motion states.
Signal B has Bsig=2.40 MHz, so β≈2.70. Its components can be separated by several coherence-bandwidth units. The channel can therefore impose different complex gains across the waveform, making a frequency-selective model and suitable equalization or multicarrier reasoning relevant.
This does not calculate error rate, throughput or equalizer performance. Those depend on signal-to-noise ratio, modulation, coding, channel estimation, interleaving, implementation and traffic. Coherence comparison classifies one propagation relationship; it is not an end-to-end service score.
Inspect the channel transfer function over the complete 2.40 MHz span. If the response contains several resolvable variations, report their scale and time. If it remains unusually smooth despite the ensemble Bc estimate, explain the difference between this realization and the statistical summary.
CHAPTER 9 OF 32 · Audit the measurement
9. Connect frequency variation to multipath delay
Multipath components arriving at different delays create frequency-dependent interference. A channel impulse response or power-delay profile preserves this delay structure; a Fourier relationship connects it with frequency response. Longer significant delay spread generally corresponds to finer frequency variation and a smaller coherence scale.
Rules of thumb sometimes express Bc as inversely related to RMS delay spread, but their constants depend on the correlation criterion and channel model. Do not treat one reciprocal formula as an exact identity. Name the convention and validate it against measured correlation when the decision matters.
NTIA ITS propagation reports analyze impulse responses, delay statistics and correlation bandwidth together. That pairing is evidence of connected quantities, not permission to derive an exact Bc from any delay number without the report’s definitions and processing.
CHAPTER 10 OF 32 · Audit the measurement
10. Check whether the sounder can resolve the channel
A measurement system has finite bandwidth, frequency step, delay resolution, dynamic range and observation time. These limits can blur closely spaced paths, miss weak late energy or smooth frequency variation. The resulting coherence estimate belongs to the sounder and processing configuration as well as the environment.
NTIA’s best-practices guide compares continuous-wave, direct-pulse, frequency-domain, correlation-based and sliding-correlator architectures. Some preserve multipath time information and estimate RMS delay spread; others serve different tasks. Select evidence produced by a sounder capable of the claimed quantity.
Record calibration, antennas, cables, transmit waveform, receiver filtering, sampling grid and noise treatment. A high-precision threshold crossing derived from coarse frequency steps or unresolved delay structure is a presentation artifact, not increased channel knowledge.
CHAPTER 11 OF 32 · Audit the measurement
11. Keep time, position and motion in scope
Coherence bandwidth can change when the transmitter, receiver or surrounding scatterers move. A corridor, open plaza and vehicle route can have different power-delay profiles. Even at one location, opening a door or moving people can alter multipath over time.
Define the stationarity region or averaging window used. If correlation is pooled across distinct environments, the estimate can describe neither environment well. Stratify by route segment or condition when the purpose is a local system decision, and report how many independent realizations support the estimate.
Do not confuse coherence bandwidth with coherence time. The former concerns frequency separation; the latter concerns time variation. Motion can influence both, but their axes and measurements differ. A channel may be frequency-flat across a narrow signal yet change rapidly over time.
Re-evaluate the fictional table with a 0.9 threshold. The crossing lies between 0.20 MHz at correlation 0.92 and 0.50 MHz at 0.76. Linear interpolation requires a drop of 0.02 out of 0.16, or one eighth of the 0.30 MHz interval, giving about 0.2375 MHz.
Signal A’s 0.20 MHz span is now about 0.84 of Bc rather than 0.22. The underlying measured table has not changed; the summary changed because the similarity criterion became stricter. Every comparison should therefore carry the threshold.
This changed case is a transfer test. It prevents learners from treating Bc as an instrument reading independent of definition. In a real design, choose the correlation or variation tolerance that connects to the model and performance question, then preserve it in reports and tests.
Suppose a second location produces correlation 0.62 at 0.20 MHz and 0.38 at 0.50 MHz. Under the 0.5 rule, linear interpolation places Bc near 0.35 MHz. Signal A now has β≈0.57 and Signal B β≈6.86. The model risk changes without altering either waveform.
The new estimate does not prove the second location is universally worse. It indicates faster frequency decorrelation for the stated measurement and threshold. Path loss, noise, interference and time variation are separate axes that may move differently.
When comparing sites, hold sounder bandwidth, antenna configuration, processing, correlation definition and motion state as consistent as practical. If a condition changes, state the change. A site label should not become a container for several undocumented protocol differences.
A wide multicarrier signal can be frequency-selective overall while each narrow subcarrier experiences approximately flat fading. The relevant comparison becomes subcarrier spacing or occupied subcarrier width versus the channel variation scale, together with the cyclic-prefix and delay-spread design.
Do not claim that OFDM “increases coherence bandwidth.” The propagation channel has not been widened by the waveform. The system partitions a wide signal into narrower components and adds processing suited to the channel’s frequency structure.
Inspect pilot spacing and channel-estimation assumptions. Pilots farther apart than the frequency-correlation scale may fail to track variations without a model. Again, Bc is a summary input to design reasoning, not a guarantee that a particular receiver implementation succeeds.
CHAPTER 15 OF 32 · Transfer the model
15. Separate coherence bandwidth from occupied bandwidth
Occupied bandwidth encloses a stated portion of signal power. Coherence bandwidth summarizes channel frequency correlation. A transmitter can keep the same occupied bandwidth while moving between channels with different coherence bandwidths, or the channel can remain similar while filtering changes the occupied bandwidth.
The existing specialist guide on occupied bandwidth versus out-of-band emissions protects another distinction: where wanted power lies versus unwanted emissions outside a necessary or assigned region. Neither quantity describes propagation correlation. Link the articles by their measurement object, not the shared word bandwidth.
A complete spectrum and channel report may therefore contain configured channel width, occupied bandwidth, emission-mask results, coherence bandwidth and signal-to-noise ratio. Each needs its own definition, reference point and uncertainty. One cannot be substituted because another is easier to measure.
Replace “The signal bandwidth is below 1 MHz, so the channel is flat” with “Under the stated 0.5 correlation definition, the measured channel has Bc≈0.89 MHz; the 0.20 MHz signal is narrow relative to that scale, so a flat-response model is a hypothesis to verify across its band.”
Replace “Delay spread equals the inverse of coherence bandwidth” with “Delay dispersion and frequency correlation are related through the channel response; any reciprocal approximation needs its model and threshold.” This wording preserves the physical link without inventing an exact conversion.
Avoid reporting a site-wide Bc from one time and position. Name route, antenna state, band, averaging window and uncertainty. If a percentile across locations is used for design, report the distribution and percentile rather than presenting the selected conservative value as every location’s channel.
Is coherence bandwidth allocated by a regulator? No. It is a propagation-channel summary; allocations and necessary or occupied bandwidth concern spectrum use. Can a narrowband signal still fail? Yes. Noise, interference, deep flat fading, timing, protocol and wider-service dependencies remain.
Does larger Bc always mean better service? It indicates slower frequency variation under the definition, not path gain, capacity or reliability. Can two studies report different Bc for the same raw channel? Yes, if they use different correlation thresholds, averaging or estimators.
Is coherence bandwidth constant with carrier frequency? Do not assume so. Propagation, antennas, environment and measurement band matter. Measure or use a justified channel model for the actual setting, and state when a value has been transferred from another band.
CHAPTER 18 OF 32 · Report and continue
18. Sources, sibling routes and a final transfer calculation
Use NTIA ITS Best Practices for Radio Propagation Measurements for channel-sounder architectures, impulse-response measurements and reported channel quantities. The ITS report on impulse responses in large outdoor cells connects delay statistics, multipath and correlation or coherence bandwidth in measured environments.
Continue with Separating Occupied Bandwidth from Out-of-Band Emissions, Separating Channel Width from Delivered Throughput, Separating Packet-Delay Variation from Clock Drift and Why the Same Bit-Error Rate Can Produce Different Packet-Error Rates.
Final transfer test: a new channel has Bc=0.40 MHz under a documented 0.5 rule. Compare signals of 0.05 MHz and 1.60 MHz: β=0.125 and 4.0. State the provisional flat or frequency-selective model for each, then list the response plot, uncertainty, time and location evidence needed before making a performance claim.
CHAPTER 19 OF 32 · Report and continue
19. Match frequency step to the threshold decision
A sounder sampled the fictional correlation curve every 0.50 MHz, so the original 0.5 crossing could only be bracketed between adjacent samples before interpolation. If the engineering decision distinguishes 0.80 from 1.00 MHz, that grid is too coarse to support the distinction without a justified response model.
Choose frequency spacing from the smallest variation that the study needs to resolve, then check that the total sounding span still covers the delays of interest after the relevant transform. Finer spacing and wider span serve different measurement roles; changing one does not automatically improve every aspect of the impulse-response estimate.
Preserve windowing, averaging and calibration choices with the grid. A smooth curve from dense plotting points may still come from coarse independent measurements. Report the measured separations, the interpolation or fitted model and the resolution at which the threshold was actually located.
Repeated channel records will not produce identical correlation values. For each frequency separation, estimate variability with a method suited to the sampling and stationarity assumptions. Then propagate that uncertainty to a bracket or interval for the threshold crossing rather than attaching several unsupported digits to Bc.
Suppose the 0.5 crossing is estimated at 0.89 MHz but plausible resampling outcomes range from 0.75 to 1.05 MHz. Signal A at 0.20 MHz remains narrow relative to every value in that interval, while a 0.90 MHz signal lies near the boundary. The second classification deserves direct response inspection instead of a binary label.
Uncertainty also includes calibration, finite bandwidth, noise-floor treatment and environmental variation. A statistical interval over repeated records does not automatically cover those systematic effects. List the included sources and keep model sensitivity, such as the chosen correlation threshold, visible.
A route survey can produce one coherence-bandwidth estimate for each local stationary segment. Summarising all segments with a mean may hide a lower tail in which the channel changes much faster with frequency. For a conservative design, a stated percentile can be more relevant than the average, but it must be named as a population summary.
Keep the paired evidence: location or condition, estimate, threshold, confidence or quality flag and signal bandwidth considered. Then calculate the ratio Bsig/Bc for each segment or for a deliberately selected percentile. Do not divide by an average Bc and claim that every point on the route has that ratio.
If some segments have no reliable crossing inside the measurement span, retain them as censored or unresolved cases under a stated rule. Dropping them can bias the reported distribution toward easy measurements and create a false sense of uniform frequency correlation.
CHAPTER 22 OF 32 · Report and continue
22. Audit the flat-fading model before using it
For a proposed flat model, plot complex H(f) across the exact occupied band for representative records. Check magnitude variation, phase behaviour, deep notches and whether the averaging used to compute Bc erased features that the receiver still experiences. The summary ratio is a screening tool, not the final model validation.
Then compare a frequency-selective model using a performance-relevant diagnostic, such as prediction residuals or validated link simulation under the authorised study design. Added complexity is useful only when it explains evidence that matters. A wide signal does not compel one specific equalizer or waveform.
Record the model’s domain: carrier band, antenna arrangement, environment, motion state, signal span, threshold convention and observation period. A later team can then tell whether reusing the model is a supported transfer or merely a familiar label applied to a new channel.
CHAPTER 23 OF 32 · Report and continue
23. Separate measurement bandwidth from signal bandwidth
The sounder bandwidth is the span over which the channel measurement can resolve delay and frequency behaviour. The operational signal bandwidth is the waveform span being classified. If the sounder covers only part of the signal band, a measured correlation scale cannot certify unobserved frequencies.
Conversely, a sounder may be much wider than the deployed signal. That can provide useful channel context, but the flat or selective decision must still inspect the sub-band actually occupied by the signal. Frequency dependence can vary across the larger band, so a centre-frequency label is insufficient.
State sounder span, usable calibrated span, operational signal span and their alignment. Exclude edge regions where filters or calibration are unreliable. This four-part ledger prevents a broad instrument setting from being confused with evidence over the intended waveform.
A new record yields a 0.5-correlation crossing bracket of 0.60 to 0.80 MHz. Signal C occupies 0.10 MHz, Signal D occupies 0.70 MHz and Signal E occupies 2.00 MHz. Their ratio ranges are about 0.125–0.167, 0.875–1.167 and 2.5–3.33 respectively.
Signal C is consistently narrow relative to the bracket, so a flat model is a strong starting point for this criterion. Signal E is consistently wide, so frequency selectivity deserves explicit modelling. Signal D straddles the boundary: the honest result is sensitivity, followed by direct H(f) inspection and performance evidence.
Now transfer the signals to a moving route where the crossing distribution has a tenth percentile of 0.25 MHz. Recompute before making a route-wide statement. The waveforms did not change, but the population of channels did; coherence bandwidth belongs to the measured channel context, while signal bandwidth belongs to the waveform.
CHAPTER 25 OF 32 · Report and continue
25. Keep coherence bandwidth separate from the receiver filter
A receiver filter can narrow the observed noise and signal spectrum, but it does not alter the underlying propagation channel’s frequency correlation. It changes which part of that channel reaches later processing. If the filter is included in the measured transfer function, say so and distinguish composite response from over-the-air channel response.
For a narrow receiver setting, an apparently flat measured band may simply be too small to reveal the crossing. Report Bc as greater than the calibrated observation span under the stated threshold, rather than assigning the span itself as the coherence bandwidth. A lower bound preserves the real information without inventing a crossing.
When comparing receiver modes, use a common calibrated channel estimate or account explicitly for each filter response. Otherwise, a smoother trace in the narrower mode can be misread as an environmental improvement. The channel, signal and instrument each contribute a different frequency scale to the evidence.
CHAPTER 26 OF 32 · Report and continue
26. Check correlation magnitude and complex correlation
A study may threshold the magnitude of frequency correlation, its real part or another normalized statistic. These choices are not equivalent when phase varies. Name the definition and preserve the complex response processing that leads to it before comparing Bc across studies.
For the fictional example, the table uses normalized magnitude correlation. A complex-correlation criterion could cross earlier because phase rotation reduces the real component even while magnitudes remain related. The difference reflects the selected similarity question, not a contradiction in the channel.
Choose the statistic that supports the intended model and performance question, then apply it consistently across locations and signals. Do not combine a Bc estimated from magnitude correlation with a design rule derived for a different convention without checking the derivation.
CHAPTER 27 OF 32 · Report and continue
27. Avoid treating one deep notch as the whole channel
A local spectral notch can strongly affect a signal even when the broader correlation curve suggests a large coherence scale. Conversely, a narrow notch should not automatically define a tiny site-wide Bc if the estimator and engineering question summarize typical correlation. Inspect both local response and the chosen statistic.
Report notch depth, width, location and persistence where they matter. Compare them with the operational signal’s occupied sub-bands and receiver processing. Frequency hopping or allocation changes can move the waveform relative to the notch without changing the environment’s underlying multipath mechanism.
This layered report prevents two errors: declaring a channel flat from one summary while hiding a critical notch, and declaring the entire band unusable from one local feature. The model should preserve the structure required by the decision.
CHAPTER 28 OF 32 · Report and continue
28. Create a reproducible channel-comparison record
Store link endpoints, antenna heights and orientations, route or location, time and motion state, carrier band, sounder span and step, calibration, averaging, correlation definition, threshold, crossing method and uncertainty. Pair Bc with the exact signal-bandwidth convention used in the comparison.
Include the correlation curve or sufficient sampled values, not only the final crossing. A later reader can then change the threshold or test interpolation sensitivity without rerunning an irreproducible measurement. Flag records where no crossing occurs inside the calibrated span.
End with the provisional model, its validation evidence and its limits. “Frequency-flat for this 0.20 MHz signal under the stated conditions” is transferable because it names the waveform and context. “The channel is flat” discards the very comparison that made the conclusion meaningful.
CHAPTER 29 OF 32 · Report and continue
29. State what would falsify the chosen channel model
A flat-fading hypothesis should come with a failure test: for example, complex-response variation across the occupied band exceeds a stated tolerance or a frequency-selective model materially reduces validated residuals. Without a falsifier, the label can survive contrary evidence by changing its meaning.
For a frequency-selective hypothesis, specify the resolution and observation conditions needed to see the structure. If the sounder or receiver filtering smooths variations below that scale, absence of visible selectivity is inconclusive. Measurement capability belongs in the test definition.
Repeat the check when signal bandwidth, carrier band, antennas, motion or environment changes. Coherence bandwidth is valuable because it organizes this transfer, but each new context must still meet the assumptions behind the model and its evidence threshold.
CHAPTER 30 OF 32 · Report and continue
30. Relate coherence bandwidth to equalizer evidence carefully
A signal wider than Bc suggests that different spectral components may experience different complex gains. It does not specify the equalizer length, pilot pattern or achievable performance. Those require a sampled channel response, delay support, noise conditions and an implementation model.
Compare a flat-gain receiver model with a frequency-resolved model on held-out channel records. If the richer model improves a decision-relevant metric consistently, that supports its use. If it merely fits measurement noise, the coherence comparison has not justified the added complexity.
For a changed waveform, recompute the occupied span and subcarrier or symbol structure before reusing the conclusion. The same physical channel can be approximately flat for one waveform and selective for another. Equalization is a system response to that relation, not a property contained in Bc alone.
Frequency correlation need not fall smoothly with separation. Multipath structure, finite samples and estimator noise can create lobes or multiple threshold crossings. Define whether Bc uses the first crossing, an envelope, an integrated measure or a fitted model, and report that rule with the value.
For a first-crossing rule, preserve the sampled curve around every early dip and its uncertainty. A single noisy point below 0.5 followed by recovery may require more records or smoothing that was specified in advance. Removing it only because it changes the answer is not a defensible estimator.
If two studies use different crossing rules, compare the curves and definitions before comparing their headline bandwidths. The disagreement may come from summary construction rather than the environment. A reproducible curve lets the reader translate between conventions when the evidence supports it.
CHAPTER 32 OF 32 · Report and continue
32. Transfer across carrier bands without silent scaling
The same room measured at two carrier bands can have different antenna patterns, material interactions, available bandwidth and resolved multipath. Do not scale Bc by the carrier-frequency ratio unless a validated model for that setting supplies the relation. Shared geometry alone is insufficient.
Use comparable antenna placement and processing where possible, then report band-specific delay profiles, frequency correlation and threshold crossings. If hardware forces different sounding spans or steps, treat that as a measurement difference and test whether both instruments resolve the variation needed for the comparison.
When a design moves bands, recompute Bsig/Bc with the new waveform and channel evidence. Preserve the old result as a bounded prior, not as a renamed measurement. This transfer check keeps propagation, waveform and instrument changes visible in the final model choice.
If only legacy data exist, report the transferred estimate as an assumption and plan a measurement that could overturn it. Compare predicted response variation with pilot or sounding evidence in the new band, and retain discrepancies rather than tuning the threshold until labels agree. The useful output is a model with a stated domain and test, not a universal coherence number attached to the site. Have an independent reader reproduce the crossing and signal ratio from the saved curve; if they cannot, the report lacks a definition, grid, threshold or unit needed for transfer.
Contents · Previous chapter · Return to Wireless Communication