Diffraction, reflection and scattering are linear operations on the propagating field. The loss imposed by a given bounce — a wall reflection, a rooftop diffraction, a scatter off a parked vehicle — is fixed in dB by the geometry, materials and wavelength involved. It does not change with the power of the signal that enters it. In a passive, non-saturated environment, path loss adds in dB regardless of transmit power: raising transmit power by 5 dB raises received power by exactly 5 dB, nothing more and nothing less.

This matters because it removes a tempting but incorrect assumption from the planning process: there is no power level at which a given bounce becomes easier to get through. The standard reference for modelling this class of short-range, sub-6 GHz, urban NLOS link is ITU-R Recommendation P.1411, which provides site-general and site-specific methods (including one-turn and two-turn street geometries) for predicting exactly this kind of multi-bounce path loss.

3. What power actually buys: link margin and the bounce budget

What transmit power does control is how many cascaded bounce losses a link can absorb before the received signal falls below the receiver's sensitivity floor — a fixed threshold set by the radio's noise figure and required SNR, independent of transmit power. Each additional bounce (or additional turn in NLOS terminology) spends more of that fixed budget. A higher transmit power simply provides a larger budget to spend, which permits more bounces — or a longer, more indirect route — before the link becomes unusable.

The table below illustrates the effect using representative, illustrative figures: a first-bounce path loss of 95 dB, a further 15 dB per additional turn, and a receiver sensitivity of -110 dBm. These specific numbers are for illustration only — real per-turn losses and achievable margins must come from site-specific modelling (ITU-R P.1411) or field measurement at the actual deployment, not assumed.

Two things follow from the table. First, the loss per turn is identical in every column — the physics of the bounce is unchanged by power. Second, the number of turns that remain usable grows with power: at 9 dBm the link tolerates two turns, at 20-27 dBm it tolerates three. This is a link-budget effect, not a change in attenuation.

4. The hidden cost: delay spread and multipath complexity

The signal arriving at the receive antenna is not one path but a sum of multipath components, each with a different delay and a different cumulative bounce loss. Components that travel further and involve more turns generally arrive later and weaker, so the receiver's sensitivity floor acts as a cutoff on how much of that power-delay profile is actually visible. At low transmit power, only the earliest and least-attenuated components clear the floor and the channel looks simple. Raising transmit power lifts the entire profile, so later, weaker, more circuitous components that were previously buried in noise begin to contribute measurable energy.

The consequence is that RMS delay spread and multipath complexity increase with transmit power, not decrease. A wider delay spread narrows coherence bandwidth, increases the likelihood that a newly-revealed weak component lands out of phase with the dominant path and carves a deep narrowband fade, and increases run-to-run RSSI variability. In short: extra power buys link margin, it does not make the channel cleaner — in some instances it makes received signal quality less predictable, since more paths are now available to interfere both constructively and destructively.

5. The high-density complication: shared spectrum and the near-far problem

In a dense deployment — many devices sharing the same channel or band — the choice of transmit power is a network decision, not just a per-link one. Raising an individual device's power to cure a marginal NLOS link does not only help that device: it raises the effective interference floor for every other device and gateway sharing the same spectrum, worsens the near-far problem for weaker neighbouring links, increases the probability of collisions in contention-based or listen-before-talk access schemes, and reduces spatial reuse of the channel across the deployment.

This is compounded by the delay-spread effect above: a device running at higher power is not just louder, it is also occupying more of the shared channel's time-frequency space with a richer, longer multipath tail. Setting every device in a dense network to maximum power to guarantee individual link closure therefore tends to reduce aggregate network capacity even as it improves any single link. The correct power for a high-density deployment is the minimum that closes each device's link with adequate margin — not the maximum available.

6. Regulatory ceiling: the 915-928 MHz ISM band in Australia

922 MHz falls within the 915-928 MHz band regulated in Australia under the ACMA Radiocommunications (Low Interference Potential Devices) Class Licence 2025, which took effect on 1 October 2025 and replaced the 2015 instrument. Under this class licence, permitted EIRP depends on the modulation scheme: non-frequency-hopping transmitters are limited to 3 mW EIRP, while frequency-hopping and digital-modulation transmitters (the category most sub-GHz IoT radios fall into) are permitted up to 1 W (30 dBm) EIRP.

This puts 27 dBm within about 3 dB of the regulatory ceiling for digital-modulation/frequency-hopping devices, leaving limited headroom to go further even if the link budget demanded it — and it puts 27 dBm roughly 20 dB over the limit for non-hopping transmitters, a distinction worth confirming against the specific device's modulation before assuming any of these four power levels is compliant. The class licence may also carry duty-cycle or channel-bandwidth conditions that interact with the power limit; the current instrument should be checked directly rather than assumed.

7. A practical method for choosing power in dense NLOS deployments

1.    Characterise the worst-case NLOS path between representative ground-level units and the gateway antenna, using ITU-R P.1411 multi-turn / rooftop-diffraction modelling or, preferably, field RSSI measurement across the candidate power levels, averaged over enough samples per location to average out fast fading rather than relying on single readings.

2.    Establish the link margin required above receiver sensitivity to hold an acceptable packet error rate under Rayleigh/Rician fading — typically 10-15 dB in dense NLOS streets, more if seasonal changes (foliage, parked vehicles, new construction) need to be tolerated.

3.    Select the minimum transmit power that clears the required margin at the worst-case device location, not the class-licence maximum. This preserves the shared channel for reuse across the rest of the deployment and keeps each link's delay-spread and interference footprint as small as possible.

4.    Where device locations are heterogeneous — some near clear LOS, some behind several turns — use per-device adaptive power and data-rate control rather than one fixed network-wide power, so only the genuinely marginal links spend the extra power budget.

5.    Re-validate periodically or after material changes to the urban clutter (new buildings, seasonal foliage, altered gateway siting), since the achievable bounce budget is a property of the physical environment and can shift the correct power setting over time.

8. Transmit power versus in-band interference

The preceding sections address power in the presence of noise and multipath. A related but distinct question is what happens when another signal — a co-located network, a foreign ISM-band device — occupies the same channel. There, the operative threshold is no longer carrier-to-noise but carrier-to-interference (C/I): the desired signal's strength relative to the interferer's strength at the receiver. If the interferer is independent and its power at the receiver does not depend on the desired transmitter, raising transmit power raises C/I dB-for-dB, just as it raises margin over the noise floor.

That linear C/I gain does not translate into linear signal quality. Digital demodulators exhibit a steep, threshold-like relationship between C/I and error rate — a waterfall curve — so below a required C/I margin the link is effectively unusable, and above a slightly higher margin, error rate is already negligible and further power buys almost nothing. This is the capture effect: once the desired signal clears the interferer by the modulation's required margin, the receiver locks onto and decodes the stronger signal cleanly; below that margin, both signals smear together unpredictably rather than degrading gracefully.

The table below illustrates this with representative, illustrative figures: a fixed in-band interferer at -95 dBm, a desired-signal path loss of 100 dB, and a required C/I of 8 dB. C/I itself grows in exact step with transmit power, confirming the underlying relationship is linear — but the practical outcome is a step function: 9 dBm is unusable, 14 dBm sits on the capture edge, and 20 dBm and 27 dBm are both clean with no meaningful difference between them.

The relationship breaks down under several conditions. If the interference source is reciprocal — another device in the same network — raising power to win the C/I contest reproduces the near-far arms race of Section 5, now played out against interference rather than noise. If the interferer is strong enough to drive the receiver's front end into compression or generate intermodulation products, no increase in desired-signal power helps: the limiting factor is the receiver's dynamic range, not the relative signal strengths, and the correct response is filtering, attenuation, or siting. The ACMA 915-928 MHz class-licence ceiling caps the lever regardless of link-budget need. And impulsive or duty-cycled interference does not respond to steady transmit power at all — during a burst the link is blocked no matter what, and between bursts no extra power was needed.

For a deployment where strong in-band interference is already known to be present, the recommended approach is to survey the actual interference on site (power, bandwidth, duty cycle, ideally over a full day or week), size the link budget against whichever is worse — noise floor or interference floor — and lean on processing gain, frequency agility, coding, and spatial technique ahead of raw transmit power. Power increases remain useful up to the capture threshold and the regulatory ceiling, but are not a scalable answer in a dense network sharing the same interference-prone band, and are no answer at all once the interferer is desensitising the receiver's front end.

9. Summary

Across 9, 14, 20 and 27 dBm, the attenuation imposed by any individual reflection or diffraction event is unchanged — it is a property of the environment, not the transmitter. What changes with power is the number of cascaded bounces a link can tolerate before falling below receiver sensitivity, and, as a side effect, how much of the environment's multipath structure becomes visible to the receiver. For a single marginal link, more power is a reasonable fix. For a high-density deployment sharing one band, the better default is the minimum power that reliably closes each link, supplemented by adaptive control for the harder cases, with the ACMA 915-928 MHz class-licence limits treated as a hard ceiling rather than a target. Where the impairment is in-band interference rather than distance or multipath, the same discipline applies with a twist: power gains are linear against C/I but the payoff is a threshold effect, valuable up to the capture margin and largely wasted beyond it.