Ekahau Sidekick and RSSI Offset: Physical Limits of the Method and Why Real Client Behaviour Cannot Be Fully Modelled / Habr

Abstract. This work examines the physical foundations of Ekahau Sidekick measurements and the device offset mechanism from the perspectives of antenna theory, receiver noise theory, statistical signal theory, and the IEEE 802.11 standard family. It is shown that the scalar received signal strength indicator (RSSI) offset constitutes a linear level shift and does not model the true signal-to-noise ratio (SNR) of the client device, the quadrature amplitude modulation (QAM) constellation structure, the rate adaptation algorithm, or roaming behaviour. In addition to five independent physical and systemic sources of inaccuracy, the paper addresses modeling assumptions in Ekahau with respect to multiple-input multiple-output (MIMO) gain, multipath propagation, airtime estimation, and SNR visualisation. Verified numerical error estimates for representative deployment scenarios and practical recommendations are provided.

I. Introduction

Ekahau Sidekick is widely deployed in professional Wi‑Fi site surveys as the primary measurement instrument. Sidekick 1 incorporates seven factory-aligned internal antennas and two dual-band Wi‑Fi radios; Sidekick 2 features nine custom wideband 3D antennas and four tri-band radios [1, 2]. These hardware advantages result in a systematically higher RSSI compared with a typical client device measured at the same location.

To compensate for this discrepancy, Ekahau implements a device offset mechanism — an empirical scalar shift of the raw RSSI reading:

$\mathrm{RSSI}{\rm corr} = \mathrm{RSSI}{\rm raw} - \Delta \quad$ (1)

where $\Delta$ [dB] is determined by the user from comparative measurements of Sidekick and the target client device [3].

The objective of this paper must be stated at the outset: this is not a critique of Ekahau as a tool. Ekahau is a professional and widely recognised solution for Wi‑Fi site surveys. The aim is to identify precisely what Ekahau's developers have simplified, to quantify where those simplifications introduce measurable error, and to provide guidance on how to interpret survey results within these constraints.

II. Physical Basis of Sidekick Superiority: Effective Aperture and Received Power

II-A Effective Antenna Aperture

The relationship between the physical characteristics of an antenna and its received power is described through the effective aperture $A_{\rm eff}$ [m²]. For an antenna with gain and operating wavelength $\lambda$ :

$A_{\rm eff} = \frac{\lambda^2}{4\pi} G \quad$ (2)

For an incident power density [W/m²], the received power is:

$P_r = S_i \cdot A_{\rm eff} = S_i \cdot \frac{\lambda^2}{4\pi} G \quad$ (3)

II-B Multipath Environment and the Angular Power Spectrum

In a real multipath propagation environment, the signal arrives from a continuum of directions. The directional power distribution is characterised by the Angular Power Spectrum (APS) $\mathcal{P}(\theta,\phi)$ [W/sr]. The received power for an antenna with normalised radiation pattern $D(\theta,\phi)$ is:

$P_r = \int_{0}^{2\pi}\int_{0}^{\pi} \mathcal{P}(\theta,\phi) \frac{\lambda^2}{4\pi} D(\theta,\phi) \sin\theta d\theta d\phi \quad$ (4)

Sidekick, with its broad and uniform radiation pattern provided by an ensemble of 7–9 optimally oriented antennas, integrates power over a larger solid angle. At the same field $\mathcal{P}(\theta,\phi)$ :

$P_r^{\rm (EK)} > P_r^{\rm (Phone)} \quad$ (5)

since $D_{\rm EK}(\theta,\phi)$ is more uniform over the sphere than $D_{\rm Phone}(\theta,\phi)$ — the radiation pattern of a smartphone with an asymmetrically positioned antenna.

III. From Received Power to SNR: The Role of Receiver Noise Figure

III-A Receiver Noise Model

The thermal noise power at the demodulator input over bandwidth $\Delta f$ [Hz] is:

$N = k T_{\rm sys} \Delta f \quad$ (6)

where $k = 1.38\times10^{-23}$ J/K is Boltzmann's constant. The system noise temperature is:

$T_{\rm sys} = T_{\rm ant} + T_{\rm receiver} = T_{\rm ant} + (F-1) T_0 \quad$ (7)

where is the noise figure of the receiver chain and T = 290 K is the standard reference temperature.

III-B SNR at the Demodulator Input

$\mathrm{SNR} = \frac{P_r}{N} = \frac{S_i\cdot\tfrac{\lambda^2}{4\pi} G}{k (T_{\rm ant}+(F-1) T_0) \Delta f} \quad$ (8)

Expression (8) makes explicit that SNR is determined jointly by the effective aperture $A_{\rm eff}$ and the noise figure . The claim $\mathrm{SNR}_{\rm EK}>\mathrm{SNR}_{\rm Phone}$ holds only when is comparable across both devices. If the client chipset exhibits a lower , the aperture advantage of Sidekick may be partially offset.

III-C Why RSSI Offset Does Not Model SNR

The RSSI offset (1) shifts by a constant $\Delta$ but does not alter — the noise floor of the target device:

$\mathrm{SNR}_{\rm corr} = \frac{P_r^{\rm (EK)}-\Delta}{N^{\rm (EK)}} \neq \frac{P_r^{\rm (Phone)}}{N^{\rm (Phone)}} \quad$ (9)

since $N^{\rm (EK)}\neq N^{\rm (Phone)}$ when and $T_{\rm ant}$ differ between devices. Consequently, the corrected RSSI value is not an estimate of the client SNR.

IV. From SNR to QAM Constellation Structure

IV-A Additive White Gaussian Noise (AWGN) Channel Model for an Orthogonal Frequency-Division Multiplexing (OFDM) Subcarrier

For a single OFDM subcarrier with suppressed inter-symbol interference (ISI) — cyclic prefix length exceeding the channel delay spread — the standard AWGN channel model is:

$y_k = \sqrt{\mathrm{SNR}}\cdot x_k + n_k,\quad n_k\sim\mathcal{CN}(0,1) \quad$ (10)

where $x_k\in\mathcal{A}$ is a symbol drawn from the -QAM alphabet $\mathcal{A}$ and is complex Gaussian noise.

IV-B Root-Mean-Square Constellation Scatter

The root-mean-square constellation scatter — the deviation of received symbols from their ideal lattice positions, also quantified by the Error Vector Magnitude (EVM) metric — is:

$\sigma_{\rm symbol} = \frac{1}{\sqrt{\mathrm{SNR}}} \quad$ (11)

Substituting (8):

$\sigma_{\rm symbol} = \left(\frac{k T_{\rm sys} \Delta f}{S_i\cdot\tfrac{\lambda^2}{4\pi} G}\right)^{1/2} \quad$ (12)

It follows directly from (12) that $\sigma^{\rm(EK)}_{\rm symbol} < \sigma^{\rm(Phone)}_{\rm symbol}$ — the QAM constellation observed by Sidekick is more compact at equivalent field conditions.

IV-C Symbol Error Rate

For -QAM in an AWGN channel, the symbol error rate (SER) is [4]:

$P_s \approx 4\left(1-\frac{1}{\sqrt{M}}\right)Q\left(\sqrt{\frac{3 \mathrm{SNR}}{M-1}}\right) \quad$ (13)

where $Q(\cdot)$ denotes the Q-function. Since the scalar offset does not restore the true SNR of the client device (9), it does not permit a correct estimation of — and therefore of the actual link quality at that location.

Caveat. The AWGN model (10) is an approximation. In a real Wi‑Fi channel, residual ISI, carrier phase errors, and power amplifier non-linearities introduce additional distortions not captured by equations (11)–(13). Nevertheless, the AWGN model provides a valid qualitative framework for comparative analysis.

V. Five Independent Sources of Client Model Inaccuracy

Reason I. Sidekick Is a Passive Sensor — It Does Not Participate in Rate Adaptation

In passive survey mode, Sidekick operates exclusively as a listener: it decodes beacons and management frames, measures RSSI and SNR, but does not associate with an access point (AP) [5].

Two distinct operating modes must be distinguished. In passive survey mode, Sidekick listens to the air without AP association. In active survey mode, a real test device — a smartphone or laptop — is connected to Sidekick, associates with the AP, and generates traffic; Ekahau is then capable of visualising with which AP the client was associated at each point along the walk path. In other words, Ekahau can show roaming behaviour — but only for the specific device used in that survey.

The fundamental constraint is that the data always reflects the roaming behaviour of the device with which the survey was conducted. Rate adaptation algorithms (Minstrel, Intel Wireless Adaptation Technology, Qualcomm-proprietary implementations, etc.) are vendor-specific and are not generalised by Ekahau. Data collected with one device cannot be directly applied to a client with a different chipset or firmware.

Reason II. The RSSI Difference Between Devices Is Non-Linear

The discrepancy in RSSI readings between two devices is non-linear with respect to signal level [6]. At −45 dBm, two devices may differ by 2 dB; at −60 dBm, by 4 dB; at −80 dBm, by as much as 10 dB.

Let the true inter-device difference be a function of level: $\delta(\mathrm{RSSI})$ . The scalar offset $\Delta = \mathbb{E}[\delta]$ is then the expectation of this function averaged over the signal range used during calibration. The systematic error at an arbitrary operating level is:

$\varepsilon(\mathrm{RSSI}) = \delta(\mathrm{RSSI}) - \Delta \quad$ (14)

For typical non-linearity of $\delta(\cdot)$ , the function $\varepsilon(\cdot)$ ranges from −2 to +8 dB across the operational RSSI range, causing the coverage boundary on the corrected heatmap to be systematically displaced — with both the sign and magnitude of the displacement depending on the local signal level.

Reason III. Orientation-Dependent Radiation Pattern of Smartphone Antennas

Sidekick is held in a fixed orientation with a quasi-isotropic radiation pattern. A real smartphone is held by the user in an arbitrary and continuously varying orientation. The user's hand acts as a dielectric screen with $\varepsilon_r \approx 50$ and $\sigma \approx 1$ S/m, causing substantial reactive near-field coupling loss at the antenna.

Experimental studies confirm that smartphone orientation and source position have a significant effect on the accuracy of RSSI measurements [7, 8]. In practice, this produces a variation in the effective gain $G(\theta,\phi)$ of a smartphone of 5–10 dB within the same spatial point, depending solely on how the user holds the device — an effect that is fundamentally uncompensatable by a scalar offset $\Delta$ .

Reason IV. Chipset Noise Figure Differences: Different SNR at Identical RSSI

From expression (8), two devices with equal received power but different noise figures $F_1 \neq F_2$ exhibit different SNR values:

$\frac{\mathrm{SNR}1}{\mathrm{SNR}2} = \frac{T{\rm ant}+(F_2-1) T_0}{T{\rm ant}+(F_1-1) T_0} \quad$ (15)

The noise figure difference between budget and flagship Wi‑Fi chipsets (Qualcomm FastConnect, MediaTek Dimensity, Apple W-series) is typically 3–5 dB. At dB with $T_{\rm ant} \approx T_0$ , the SNR ratio from (15) is approximately 2 — meaning that two devices with identical RSSI may differ in actual SNR by a factor of two. The scalar offset operates exclusively on and has no mechanism to account for of the target device.

Reason V. Roaming Behaviour: Device-Specific Handover Thresholds

Wi‑Fi engineers understand from first principles that roaming occurs under different conditions for different clients. Ekahau permits the user to configure calculated cell overlap zones when the target client's characteristics are known in advance — however, this is a calculated model, not a measurement of the behaviour of a specific device.

According to Apple's official documentation, iPhone and iPad maintain association with the current AP until the RSSI drops below −70 dBm; Mac computers use a threshold of −75 dBm [10]. Android-based industrial scanners with conservative roaming algorithms may retain association significantly below the roaming threshold typical of smartphones — a well-known sticky client phenomenon.

Consequently, even with configured overlap zones, Ekahau cannot predict the roaming behaviour of a device that was not used in the survey. Verification of roaming for a specific client requires an active survey conducted with that exact device [9].

VI. Structural Simplifications in Ekahau Beyond the RSSI Offset

VI-A MIMO Gain Is Not Modelled

The actual channel capacity under spatial multiplexing is determined by the rank structure of the channel matrix $\mathbf{H}$ :

$C = \sum_{i=1}^{\min(N_t,N_r)}\log_2\left(1+\frac{\lambda_i^2\cdot P}{N\cdot N_t}\right) \quad$ (16)

where $\lambda_i$ are the singular values of $\mathbf{H}$ , and , are the numbers of transmit and receive antennas respectively. MIMO gain is a function of spatial correlation between antenna elements — itself a function of the propagation geometry, angular spread, and the relative positioning of AP and client. Ekahau does not model channel rank and does not account for MIMO gain in its coverage heatmaps.

VI-B Multipath Propagation and Reflections Are Not Modelled

Ekahau constructs coverage maps using an attenuation-based propagation model that accounts for wall losses. The real channel is formed by the superposition of multipath components, each carrying its own delay, phase, and amplitude. This produces frequency-selective fading that governs the SNR variation across OFDM subcarriers. In environments with pronounced multipath, the actual per-subcarrier SNR may deviate substantially from the mean in either direction — a phenomenon invisible to Ekahau's model.

VI-C Airtime Estimation Is Constrained by Model Accuracy

A correct airtime estimate requires knowledge of each client's modulation and coding scheme (MCS), MAC protocol data unit (MPDU) size, aggregated MPDU (A-MPDU) aggregation behaviour, protocol overhead, and the number of active stations. Ekahau models airtime from attenuation libraries and configured client profile parameters; the accuracy of this model is directly contingent on the quality of calibration and the completeness of survey data. In dense networks with high MCS variability across clients, the modelled airtime estimates may deviate substantially from measured values.

VI-D The Displayed SNR Reflects Sidekick's Noise Floor, Not the Client's

Ekahau Sidekick measures the noise floor at the point of measurement via integrated spectral analysis — a more accurate approach than a fixed constant. However, the displayed SNR reflects the noise environment as observed by Sidekick, not the noise floor of the specific client chipset. The difference between these quantities is governed by the noise figure discrepancy expressed in (7). In saturated RF environments, this difference may amount to several decibels, producing a systematic offset between the displayed SNR and the actual SNR experienced by the client device.

VII. Magnitude of Real-World Error: Verified Scenarios

The following case studies quantify the error magnitudes, with all numerical values verified against standard IEEE 802.11 reference tables.

Scenario A. Coverage Boundary Prediction Error at −78 dBm

At a signal level of −78 dBm, the typical error of a scalar offset due to non-linearity (Reason II) is 6–8 dB. The signal attenuation gradient is strongly environment-dependent: approximately 0.5–1 dB/m in open-plan offices and 1.5–3 dB/m in corridors with partitions.

With an error of 6–8 dB at a gradient of 1.5 dB/m, the coverage boundary prediction error amounts to 4–5 metres. At 3 dB/m, the prediction error is 2–3 metres. Given typical inter-AP spacings of 10–20 metres, such boundary displacement directly affects decisions regarding AP count and placement.

Scenario B. MCS Selection and Actual Throughput at the Coverage Edge

Consider a client operating on a 20 MHz channel (802.11ac, 1 spatial stream). The SNR thresholds and physical layer (PHY) rates from the standardised 802.11ac reference table [11] are:

Very High Throughput (VHT) MCS	Modulation	Coding Rate	Min. SNR [dB]	PHY Rate [Mbit/s]
MCS 5	64-QAM	2/3	18	52
MCS 6	64-QAM	3/4	20	58.5
MCS 7	64-QAM	5/6	25	65
MCS 8	256-QAM	3/4	29	78

Table caveat. The SNR thresholds listed are IEEE reference values (packet error rate, PER < 10%). Real devices employing maximum ratio combining (MRC) receiver processing frequently achieve the same MCS at SNR values 3–6 dB below the tabulated thresholds. Nevertheless, the offset error systematically biases MCS estimates and the associated throughput predictions in a predictable direction.

If the client's actual SNR is 23 dB while the offset error introduces a 3 dB systematic overestimation, Ekahau will display a level comfortably above the MCS 7 threshold (25 dB). The real client, however, operates at MCS 6 (20 dB <= SNR < 25 dB). Upon a further 3 dB channel degradation due to human body blockage or multipath reflection — the client falls to MCS 5: the expected 65 Mbit/s becomes a measured 52 Mbit/s, a 20% reduction that is systematic and cumulative in dense deployments.

For 80 MHz channels or 2 spatial streams, the MCS 8/MCS 7 boundary occurs at SNR 35/31 dB; the corresponding PHY rate differential is 390/292.5 Mbit/s — a gap of approximately 100 Mbit/s per stream under the same offset error.

Scenario C. Industrial Scanner in a Cell Overlap Zone

An industrial scanner exhibiting sticky client behaviour retains association with AP1 until the received signal level drops to −85 dBm. The Sidekick heatmap indicates a signal level of −65 dBm from AP2 at the relevant location — sufficient for a high MCS. However, the scanner will not roam to AP2 until the AP1 signal falls to its roaming trigger threshold.

Within this zone, the scanner communicates with AP1 at MCS 1–2, receiving actual throughput substantially below what the heatmap would suggest. The Sidekick map correctly indicates coverage and good SNR from AP2 — but it cannot show that the client is physically associated with AP1 and operating at a severely degraded MCS. The scalar offset has no bearing on this scenario whatsoever.

VIII. Formal Applicability Conditions for RSSI Offset

A scalar offset provides acceptable accuracy only when all four of the following conditions are satisfied simultaneously:

C1 (Range linearity). The offset $\Delta$ was measured over a range $[\mathrm{RSSI}_{\rm min},\mathrm{RSSI}_{\rm max}]$ and the operational survey zone lies entirely within this range: $\varepsilon(\mathrm{RSSI})\approx0$ .

C2 (Comparable noise figure). The noise figures of Sidekick and the target device are comparable: $|F_{\rm EK}-F_{\rm Phone}|\ll1$ dB.

C3 (Stable device orientation). The target device is operated in a fixed or deterministic orientation.

C4 (No critical roaming zones). The survey area contains no cell overlap regions that are critical for the specific target client's roaming behaviour.

In practice, all four conditions are rarely satisfied simultaneously. A typical site survey scenario — heterogeneous signal levels, diverse client chipsets, mobile users with unpredictable device orientation — systematically violates at least one of them.

IX. Practical Recommendations

IX-A Measure the offset at signal levels relevant to your deployment objective. If the coverage boundary is defined at −70 dBm, the offset must be measured at −70 dBm — not at −50 dBm close to the AP. This minimises $\varepsilon(\mathrm{RSSI}_{\rm target})$ as defined in equation (14).

IX-B Conduct an active survey with the real target device for critical clients. Only this approach yields factual data on MCS, throughput, and roaming behaviour specific to a given chipset and rate adaptation algorithm. Ekahau's active survey mode, combined with the real target device, can visualise association and roaming — this capability should be used.

IX-C Verify roaming behaviour with the target device in the field. A walk-through of the coverage perimeter with continuous monitoring of the basic service set identifier (BSSID) and MCS is the only reliable method for observing actual client roaming behaviour. Calculated cell overlap zones in Ekahau provide only an approximation based on configured parameters, not measured behaviour.

IX-D Account for orientation-dependent effects for specific device form factors. For devices with pronounced orientation sensitivity — a handset held to the ear during VoIP, a tablet in landscape orientation, or a scanner in its working position — the offset must be measured with the target device in its operational orientation.

IX-E Apply MIMO model limitations in specific propagation environments. In environments with high spatial correlation — long narrow corridors, spaces with metallic structural elements — the actual MIMO rank may be lower than the modelled value, reducing the achievable throughput relative to Ekahau's prediction.

X. Conclusion

Ekahau is a professional and widely recognised instrument whose developers deliberately simplify reality in certain respects in the interest of practical usability. Understanding these simplifications is a prerequisite for using the tool correctly.

The RSSI offset (1) is a linear level shift. It does not model: the actual SNR of the client device as a function of its specific noise figure (Reason IV); symbol error probability and achievable MCS (equations (11)–(13)); the orientation-dependent radiation pattern of the smartphone antenna (Reason III); roaming behaviour governed by device-specific roaming trigger thresholds (Reason V); or the non-linearity of the inter-device RSSI difference across the dynamic range (Reason II).

Beyond the RSSI offset, Ekahau does not account for MIMO gain, does not model multipath propagation, employs a constrained airtime estimation model, and displays SNR as observed by Sidekick rather than the SNR experienced by a specific client chipset.

The magnitude of these errors is not merely academic:

At the coverage boundary near −78 dBm, the coverage boundary prediction error amounts to 2–5 metres depending on the local signal attenuation gradient.
A 3 dB SNR overestimation at the MCS 7 threshold shifts the client to MCS 6; upon further degradation, to MCS 5: the predicted 65 Mbit/s becomes 52 Mbit/s (−20%), a systematic and cumulative effect in dense networks. In an 80 MHz channel, the analogous transition produces a gap of approximately 100 Mbit/s per spatial stream.
An industrial sticky client in a cell overlap zone may operate at MCS 1–2 in a location where the Sidekick heatmap indicates good coverage from a neighbouring AP.

Ekahau remains a practically valuable tool for rapid coverage assessment when conditions C1–C4 are met. For deployments requiring accurate prediction of specific client device behaviour, the scalar offset must be supplemented by an active survey with the real target device — and by a clear understanding of what the tool simplifies and why.

References

Ekahau Sidekick Data Sheet. Ekahau Inc., 2020.
Ekahau Sidekick 2 Data Sheet. Ekahau Inc., 2023.
ToDSFromDS. Thoughts About Offsets in Ekahau Pro, 2019. todsfromds.com
Proakis J.G., Salehi M. Digital Communications, 5th ed. McGraw‑Hill, 2008. Ch. 6.
Wi‑Fi Vitae. Data Collected During an Active Survey, 2024. wifivitae.com
ToDSFromDS. A Deep Dive Into RF Offsets, 2023. todsfromds.com
Boussad et al. Evaluating Smartphone Accuracy for RSSI Measurements. IEEE Transactions on Instrumentation and Measurement, 2021.
Boussad et al. Evaluating Smartphone Accuracy for RSSI Measurements. INRIA HAL, 2020.
Mysteries of Client Roaming Revealed. 7SIGNAL Whitepaper, 2020.
Apple Inc. Wi‑Fi Roaming Support in Apple Devices. Apple Support HT203068, 2022.
802.11n/HT and 802.11ac/VHT MCS, SNR and RSSI Reference Table. WLAN Professionals, 2014.

Terminology Notes

"Industrial scanner" — as used in this paper, a collective term denoting any corporate Wi‑Fi client with a non-standard roaming algorithm: handheld data terminals (Zebra, Honeywell, Datalogic), VoIP handsets (Cisco, Spectralink), medical devices, and POS terminals. The roaming threshold of each device class is individual and frequently configurable. The value of −85 dBm cited in the paper is an illustrative example of conservative roaming behaviour and should not be interpreted as a typical specification for the entire device category.