OFDM Cyclic Prefix Multipath ISI Case Study

This case study follows a private OFDM wireless link that had good received power and acceptable noise margin but still produced bursts of packet loss at one industrial site. The failure was not a weak-signal problem. The dominant problem was a delayed reflected path whose significant energy arrived after the cyclic prefix. The receiver equalizer could estimate the channel, but part of the previous OFDM symbol leaked into the useful symbol interval and degraded the constellation.

The case teaches a common telecommunications lesson: a link budget can close while the waveform fails. Multicarrier systems need power, SNR, synchronization, linearity, and a channel impulse response that fits inside the guard interval or is otherwise handled by the receiver design.

Case Summary

Item	Engineering relevance
System	Private OFDM radio link carrying operational telemetry and maintenance traffic across an industrial site.
Symptom	Packet loss and EVM spikes in one sector while received signal power remained healthy.
Original assumption	Add transmit power or force a lower modulation mode.
Hidden weakness	A strong delayed reflection exceeded the normal cyclic prefix.
Primary evidence	Channel impulse response showed a significant delayed tap at 6.8\ \mu\text{s} excess delay.
Consequence	Inter-symbol and inter-carrier interference reduced usable SINR below the selected 64-QAM mode.
Corrective action	Antenna relocation and pattern control, with extended-cyclic-prefix fallback as an operational contingency.

The central engineering question was:

Is the link failing because the received signal is too weak, or because the channel delay spread is incompatible with the selected OFDM guard interval?

The evidence points to the second answer.

Initial Link Data

The simplified OFDM waveform and field measurements were:

Quantity	Symbol	Value
subcarrier spacing	\Delta f	15\ \text{kHz}
useful OFDM symbol time	T_u	66.7\ \mu\text{s}
normal cyclic prefix time	T_{cp}	4.7\ \mu\text{s}
measured maximum significant excess delay	\tau_{sig}	6.8\ \mu\text{s}
delay measurement standard uncertainty	u_\tau	0.4\ \mu\text{s}
reflected tap power relative to direct path		-13\ \text{dB}
clear-channel SNR estimate		27\ \text{dB}
selected mode before fix		64-QAM, code rate 3/4
64-QAM usable SINR requirement after margin		22\ \text{dB}
service payload requirement		50\ \text{Mbit/s}
observed packet loss during events		3.5\% to 6\%

The channel sounded clean in a bench test and in a nearby open area. The problem appeared only when the antenna saw a metal-clad processing building and elevated pipe rack that created a delayed reflected path.

Step 1: Useful Symbol Time

For OFDM, useful symbol time is approximately the reciprocal of subcarrier spacing:

With:

the useful symbol time is:

The cyclic prefix is a guard interval copied from the end of the OFDM symbol. If the significant channel delay spread fits inside this interval, the receiver can often preserve subcarrier orthogonality after removing the prefix and applying frequency-domain equalization.

Step 2: Cyclic Prefix Margin

Define cyclic-prefix time margin as:

Substitute the measured values:

The guard interval is short by:

The result remains negative even if the excess-delay estimate is reduced by one standard uncertainty:

Engineering Comment

This is the first decisive calculation. The issue is not a vague “bad RF environment.” The measured channel contains significant energy outside the guard interval, so increasing transmit power will not restore the orthogonality that the OFDM receiver assumes.

Step 3: Excess Path-Length Interpretation

A delay can be translated into approximate excess path length using:

with:

For the measured delayed tap:

The normal cyclic prefix corresponds to:

So the reflected path had about:

more excess path length than the normal cyclic prefix could cover.

Engineering Comment

This path-length check makes the field evidence plausible. The site geometry included a long diagonal reflection path from a large metal structure and water-facing hard surfaces. The number is too large for ordinary indoor multipath, but plausible for an outdoor industrial radio path with distant reflectors.

Step 4: Effective Interference from the Late Tap

The delayed tap power ratio is:

Not every bit of reflected energy becomes uncancelled interference, because receiver filtering, equalization, channel estimation, and symbol timing all matter. For a screening estimate, use the fraction of the delayed path beyond the cyclic prefix:

Approximate late-tap interference ratio:

Equivalent carrier-to-late-interference ratio:

This is not a full OFDM receiver model. It is a screening calculation that shows why a moderate reflected path can dominate performance when it lies outside the guard interval.

Step 5: Combined SINR Estimate

The clear-channel SNR is:

Convert to linear form:

The late-tap interference ratio is:

An approximate combined SINR is:

In decibels:

The selected 64-QAM mode required about:

after implementation and fading margin, so the mode did not have enough usable SINR during the multipath event.

Engineering Comment

The received power looked healthy because the direct and reflected energy still reached the receiver. The problem was that part of that energy arrived at the wrong time for the OFDM symbol structure. That is why received-power indicators alone did not explain the failure.

Step 6: EVM and Service Impact

Error vector magnitude is often used as a practical modem-quality metric. A simplified relationship is:

Using the combined estimate:

So:

The field modem reported EVM bursts in the 10\% to 14\% range, which is consistent with the screening estimate. Packet loss rose because the selected modulation and coding mode was too aggressive for the distorted channel.

The operations team initially forced a lower modulation and coding mode. That reduced packet loss, but it also reduced payload capacity. A fallback-only fix was not enough because the service requirement was 50\ \text{Mbit/s} during normal operation.

Step 7: Throughput Tradeoff

Use a simplified net spectral-efficiency estimate:

where:

• k=\log_2(M) bits per symbol;
• R_c is coding rate;
• T_u/(T_u+T_{cp}) is cyclic-prefix time efficiency;
• \alpha_{oh} is pilot, control, and framing overhead.

Assume:

For 64-QAM, k=6 and R_c=3/4:

For a 20\ \text{MHz} channel:

This nominally meets the 50\ \text{Mbit/s} service requirement, but it fails in the multipath condition.

If the modem falls to 16-QAM with coding rate 1/2:

That mode is more robust but cannot carry the committed service.

Engineering Comment

Adaptive modulation is a service tool, not a substitute for fixing the channel. A fallback mode can preserve management access and telemetry, but if it drops below the committed payload requirement, the link is still degraded from an operations perspective.

Corrective Action

The engineering team used two layers of correction.

First, it changed the RF path:

• moved the sector antenna away from a reflective pipe-rack sightline;
• reduced downtilt toward the distant reflector;
• added a narrower antenna pattern;
• rechecked polarization alignment;
• repeated channel sounding at the same traffic load and time-of-day windows.

Second, it changed operating rules:

• prohibit 64-QAM when measured cyclic-prefix margin is negative;
• force extended cyclic prefix during known temporary reflector conditions, if supported;
• alarm on EVM bursts, packet-loss bursts, and channel impulse response taps outside the normal guard interval;
• preserve field channel-sounding plots in the commissioning record.

Post-Correction Validation

After antenna changes, the significant delayed tap moved below the review threshold and the maximum significant excess delay was:

New cyclic-prefix margin:

Validation results were:

Metric	Before correction	After correction	Acceptance
maximum significant excess delay	6.8\ \mu\text{s}	3.1\ \mu\text{s}	below 4.7\ \mu\text{s}
cyclic-prefix margin	-2.1\ \mu\text{s}	+1.6\ \mu\text{s}	positive margin
EVM burst range	10\% to 14\%	3.8\% to 5.2\%	below 64-QAM limit
packet loss during test window	3.5\% to 6\%	below 0.2\%	within service limit
payload throughput	unstable	68\ \text{Mbit/s} to 72\ \text{Mbit/s}	above 50\ \text{Mbit/s}
fallback events	frequent	none during acceptance window	acceptable

The link was released only after the corrected channel was validated under representative operating conditions, not merely after the antenna was moved.

Release Criteria

The final release package required:

1. received power and SNR margin above the selected mode threshold;
2. cyclic-prefix margin positive at all tested antenna azimuths and operating states;
3. EVM below the modem’s 64-QAM acceptance limit with margin;
4. packet loss below the service threshold at the committed load;
5. no unexplained fallback events during the acceptance window;
6. channel impulse response plots archived with time, antenna settings, and traffic condition;
7. monitoring thresholds for EVM, packet loss, fallback state, and late-tap alarms.

Engineering Lessons

1. OFDM links can fail from timing-domain channel shape even when received power is good.
2. A cyclic prefix is not generic overhead; it is an explicit guard interval with a physical delay-spread limit.
3. Multipath diagnosis needs channel impulse response, EVM, packet-loss, and mode-state evidence, not only RSSI.
4. Increasing transmit power does not fix energy that arrives outside the guard interval.
5. Adaptive modulation protects service only if the fallback capacity still satisfies the service requirement.

The transferable lesson is that waveform assumptions are part of the link budget. RF margin, digital demodulation, equalization, traffic requirements, and field validation must be reviewed together.