The Corner That Does Not Match the Prediction
A three-story office runs 5 GHz-only APs with radios set to 17 dBm into 6 dBi omnidirectional antennas, fed by 3 m of LMR-400 (0.22 dB/m at 5 GHz) plus 1.5 dB of connector loss. The regulatory EIRP ceiling for the band is 30 dBm. In the northeast corner, next to a concrete stairwell and a metal-sheathed file room, users drop from about 300 Mbps to 20 Mbps. The survey tool reports -71 dBm and 19 dB SNR at the worst seat, and the serving AP shows 45% retries with clients stuck at MCS 1-2. Free-space prediction for the 40 m path was -58 dBm. Budget is approved for one additional AP and cabling, but the building owner will not permit new conduit in the stairwell.
1. Classify the root cause and name the first measurement you would take. Explain why raising transmit power is or is not an appropriate response.
Read the worked answer
Evidence. Free-space prediction is -58 dBm and the measured value is -71 dBm, so about 13 dB of excess loss exists on a path that free space says is comfortable. That is obstruction loss, not distance loss.
The mismatch is the tell. At -71 dBm and 19 dB SNR in 20 MHz, a healthy link should negotiate well above MCS 1-2. Retries at 45% with low MCS means the frames are being corrupted or not heard, not that the signal is simply weak. Candidates: multipath/null from the metal wall, an elevated noise floor (-71 minus 19 implies roughly -90 dBm, which warrants checking), a non-Wi-Fi interferer, or an AP antenna pattern distorted by mounting against metal.
First measurement. A spectrum analysis at the corner plus per-AP noise-floor and retry-versus-RSSI trending, and confirmation whether the retries are counted at the AP or at the client.
Power is the wrong lever. Going 17 to 20 dBm buys 3 dB against 13 dB of obstruction, enlarges the cell and co-channel interference, and creates an asymmetric link where the client hears the AP but the AP cannot hear the client. Battery clients will not raise their power anyway. Fix the path, not the dial.
2. Compute current EIRP and EIRP at 20 dBm. What is the highest radio power that keeps this installation inside a 30 dBm EIRP limit with this antenna and cabling?
Read the worked answer
Current EIRP = 17 dBm - 0.66 dB cable - 1.5 dB connectors + 6 dBi = 20.84 dBm.
At 20 dBm radio = 20 - 0.66 - 1.5 + 6 = 23.84 dBm, still 6.2 dB below the 30 dBm ceiling.
Maximum compliant radio power = 30 dBm - 6 dBi + 0.66 + 1.5 = 26.16 dBm.
Two caveats before anyone dials it up. First, EIRP limits vary by sub-band and by regulator, and several regimes require you to reduce conducted power once antenna gain exceeds a reference value (commonly 6 dBi), so the arithmetic above must be checked against the actual rule set for the channels in use. Second, compliance is not the same as good design: 26 dBm into a 6 dBi omni in an open office will produce oversized cells, more co-channel overlap, and more sticky clients. The number answers the legal question, not the engineering one.
3. The vendor proposes swapping the 6 dBi omni for a 12 dBi patch aimed at the corner. What improves, what new risk appears, and how do you verify it?
Read the worked answer
What improves. Roughly +6 dB in the main lobe, which covers most of the 13 dB obstruction gap, and a narrower beam that reduces energy thrown at the metal wall and the stairwell, lowering multipath and stray co-channel footprint.
What gets worse. Coverage outside the main lobe drops, so seats behind or beside the patch may now be worse than before. The cell shape changes from a circle to a lobe, which invalidates the existing channel plan around it. A high-gain antenna mounted near metal can produce unpredictable pattern distortion and back-lobes. EIRP rises to about 26.8 dBm at the current 17 dBm radio setting, so re-check the regulator's antenna-gain power-reduction rule and the sub-band limit - the correct answer may be to reduce conducted power, which gives back part of the gain.
Verification. Post-install validation survey at the previously bad seats and at the seats that were fine, plus retry and MCS trending on the affected AP and its neighbors, and a check that the neighbor cells did not start showing new co-channel retries. Do not accept a predictive plot as proof.