A Straightforward Dynamic Range Error Analysis

Marion Baggett and Brett T. Walkenhorst

NSI-MI Technologies
Suwanee, GA 30024, USA
[email protected]
[email protected]

Abstract— The significant measurement standards in the antenna signal level in the pattern. An example analysis is given in
measurement community all present suggested error analysis Section VII.
strategies and recommendations. However, many of the factors in
these analyses are static in nature in that they do not vary with II. NOTATION AND BASIC MAGNITUDE MANIPULATIONS
antenna pattern signal level or they deal with specific points in the The notation used in this paper is to use “E” as the dB
pattern, such as realized gain, side lobe magnitude error or a representation of a magnitude error and “ε” as the linear
derived metric such as on-axis cross polarization. In addition, representation of a magnitude error or a phase error. These
many of the constituent factors of the error methods are the result
of analyses or special purpose data collections that may not be
equations compute the “worst case” error from the desired
available for periodic measurement. The objective of this paper is signal. In addition, all equations are simplified by considering
to use only a few significant factors to analyze the error bounds in the desired signal to have a magnitude of 1 in the linear domain.
both magnitude and phase for a given antenna pattern, for all Magnitude errors are normally represented as either a +/- delta
levels of the pattern. Most of the standards metrics are errors of in dB or a ratio in dB. In both cases, the user needs to convert
amplitude. However, interest is increasing in determining phase these magnitude errors into a linear +/- delta for eventual RSS
errors and, hence, this methodology includes phase error analysis of a composite magnitude error, which is then converted back
for all factors. into a +/- composite error in dB. When the error starts as a +/-
delta in dB, the conversion to a linear error is:
I. INTRODUCTION
Errors in a pattern measurement can be divided into four 𝐸 = 𝑝𝑟𝑜𝑣𝑖𝑑𝑒𝑑 + 𝑣𝑎𝑙𝑢𝑒 𝑖𝑛 𝑑𝐵 (1)
classifications: 𝐸 = 𝑝𝑟𝑜𝑣𝑖𝑑𝑒𝑑 − 𝑣𝑎𝑙𝑢𝑒 𝑖𝑛 𝑑𝐵 (2)
 Errors that are static, no matter the signal level. This 𝜀 = 10 −1 (3)
type of error includes mechanical pointing errors,
𝜀 = 10 −1 (4)
mutual coupling error and VSWR errors.
 Errors that vary with the signal level and are a part of Suppose an error is provided/measured/estimated at +/-0.5
the receiving system itself. The primary error source of dB. If the plus and minus linear errors are computed from just
this type is receiver linearity versus dynamic range. the power term, the results are 1.059254 and 0.944061. Since
we are looking for errors from the desired signal, the “-1” in
 Coherent errors that are static in themselves, but their equations (3) and (4) give us +0.05924 and -0.05594 as the linear
relationship to the signal vary with signal level. errors, which are peak variation about the nominal signal that we
Chamber/Range stray signals are the predominant error desire.
of this type.
For the error term provided as a ratio in dB, an additional
 Non-coherent errors whose effect varies with the signal preliminary step is required. The dB ratio must be converted
level, the most significant being thermal noise error. into a linear ratio. The +/- of that linear ratio are converted into
This paper will present the methodology of computing relative +/- dB errors and then converted to +/- relative linear
amplitude and phase errors of all four types, and will combine errors. Note that for the purposes of this study, the ratio error is
them into a composite error normalized by the peak signal in always represented as a signal to error ratio.
the pattern. This is a different error formulation than [1], [2] and 𝐸 = 𝑝𝑟𝑜𝑣𝑖𝑑𝑒𝑑 𝑟𝑎𝑡𝑖𝑜 (5)
[3], in that the standards look at specific points such as peak
signal or a side lobe level, while this analysis is creating error 𝜀 = 10 (6)
bounds that can be applied to any value in a pattern. Section II
describes the notation and basic magnitude error manipulations To convert εR into E+ and E-, the equations are different for
used throughout. Sections III through VI describe each of these coherent and non-coherent errors. For coherent errors:
errors in greater detail. The intent of the paper is to present a 𝐸 = 20 ∗ log (1 + 𝜀 ) 𝑓𝑜𝑟 𝑎𝑙𝑙 𝜀 (7)
methodology that can be used at the level of spreadsheet
capabilities to compute the expected composite error bounds for 𝐸 = 20 ∗ log (1 − 𝜀 ) 𝑤ℎ𝑒𝑛 𝜀 < 1 (8a)
any signal level in a given dynamic range. This can be applied 𝐸 = 20 ∗ log (𝜀 − 1) 𝑤ℎ𝑒𝑛 𝜀 > 1 (8b)
to experimental data to graphically show the error bounds at any
 𝐸 = −∞ 𝑤ℎ𝑒𝑛 𝜀 = 1 (8c) D. Example Receiver Linearity Error
Equations (7) and (8) are the well-known equations 14.55 Figures 1 and 2 show the linearity error versus dynamic
and 14.56 from [4]. range for a receiver with ± 0.04 dB/decade amplitude linearity
and ± 0.5 degrees of phase/decade.
For non-coherent errors:

𝐸 = 20 ∗ log (1 + 𝜀 ) 𝑓𝑜𝑟 𝑎𝑙𝑙 𝜀 (9)

𝐸 = 20 ∗ log (1 − 𝜀 ) 𝑖𝑓 𝜀 < 1 (10a)
𝐸 = −∞ 𝑓𝑜𝑟 𝜀 ≥ 1 (10b)

Note that equations (8c) and (10b) handle the case where εR
= 1. Equations (3) and (4) are then used to convert E+ and E-
into ε+ and ε- for both coherent and non-coherent errors.

III. STATIC ERROR CALCULATIONS

Static errors by their nature are constant and are applied to
any pattern signal level with the same values. They are generally
measured/specified as ± errors in dB and phase.
The formulation of the magnitude error follows equations (1) Figure 1. Receiver Amplitude Linearity Error
and (2). Therefore, the linear errors are computed with equations
(3) and (4). The phase error is already a linear value and is used
in the composite error as provided.
For example, a static error may be provided as ± 0.05 dB and
± 2 degrees in phase. Using equations (3) and (4) result in a
positive linear error of 0.005773 and a negative linear error of -
0.005740. The phase error is used as it is provided, ± 2 degrees.

IV. RECEIVER LINEARITY CALCULATIONS

A. Model
An inherent characteristic of any VNA/receiver is its
linearity. Most VNAs/receivers are specified for a linearity per
decade of the dynamic range of the instrument. The linearity is
usually formulated into ± dB / decade and ± phase degrees / Figure 2. Receiver Phase Linearity Error
decade.
V. COHERENT ERROR CALCULATIONS
B. Calculation of the Magnitude Error
An error signal is coherent when its power varies relative to
A simple approach would be to apply the specified values the desired signal power with a fixed phase offset. An example
into each decade of the receiver response. That is, the first of coherent antenna measurement errors are the stray signals
decade of values would all be ± the specification, while the from chamber absorber walls.
second decade would be the ± 2 * the specification, and so on.
However, [1] and [5] use the technique of [6] to calculate the
A. Model
errors of this type as:
Figure 3 shows a graphical model for the analysis of coherent
𝐸 = (𝐷𝑒𝑐𝑎𝑑𝑒 𝑜𝑓 𝑆𝑖𝑔𝑛𝑎𝑙 ∗ 𝐸𝑟𝑟𝑜𝑟 𝑝𝑒𝑟 𝑑𝑒𝑐𝑎𝑑𝑒)/√3 (11a) error. In all three sub-plots, the desired signal is vertical,
indicated by the solid line red arrows. The absolute phase of this
𝐸 = −(𝐷𝑒𝑐𝑎𝑑𝑒 𝑜𝑓 𝑆𝑖𝑔𝑛𝑎𝑙 ∗ 𝐸𝑟𝑟𝑜𝑟 𝑝𝑒𝑟 𝑑𝑒𝑐𝑎𝑑𝑒)/√3 (11b) signal is fixed for simplicity of illustration, but in practice can be
Equations (3) and (4) are then used to compute the ± linear anything. The principle illustrated here is valid regardless of the
errors. actual phase. The magnitude of the error signal, shown by the
dashed line blue arrows, is constant in each sub-figure and can
C. Calculation of the Phase Error be at any phase. The circle formed by these possibilities, shown
as a dashed blue line, defines the possible resultant vector
The phase calculation uses (11) to compute the plus and combinations measured by the test system, which are shown by
minus phase error entering the phase error per decade into the the dot-dash line green arrows. Note that the resultant phase will
equation instead of the magnitude error per decade. vary by the diameter of the circle relative to the magnitude of the
desired signal. The limits on the resultant magnitude occur when
the error signal’s phase relative to the desired signal is 0 or 180
degrees.
 to:
𝜀 = sin (𝜀 ) 𝑓𝑜𝑟 𝜀 ≤ 1 (12𝑎)
𝜀 = sin (−𝜀 ) 𝑓𝑜𝑟 𝜀 ≤ 1 (12𝑏)
where again we have normalized the desired signal power to
unity. Reference [8] stopped at 𝜀 = 1. The authors have added
the phase error for the stray signal larger than the signal as:
𝜀 = +180 𝑓𝑜𝑟 𝜀 > 1 (12𝑐)
𝜀 = −180 𝑓𝑜𝑟 𝜀 < −1 (12𝑑)
a) b) c) Figure 5 shows the coherent phase error versus the error ratio.
Inspection of Figures 3a and 5 shows that for a large signal to
Figure 3. Coherent Errors a) Error < Signal error ratio, the resultant phase variation is small. For the equal
b) Error = Signal c) Error > Signal ratio ±90 degrees are the asymptotic limits, while for a large
error to signal ratio, the phase can be any value, as shown in
Figure 3 a) shows the condition when the desired signal is
Figure 3c.
larger than the error signal. Figure 3 b) shows when the error
and desired signals are equal. The magnitude can range from
double the desired signal (+6 dB) to 0, which would be a -∞ dB
level. It can also be observed in the figure that the phase error
asymptotically approaches ±90 degrees. Figure 3 c) shows when
the error signal is larger than the desired signal. Again, the
resultant magnitude will have limits when the phase error is
either 0 or 180 degrees. A significant change is that the resultant
phase can be anything around the circle, making the phase error
limits ±180 degrees anytime the error is larger than the desired
signal.

B. Calculation of the Coherent Error Magnitude

Since the coherent error is usually provided as a ratio of
signal to error (5), equation (6) is used to compute the linear ratio
error. Equations (7) and (8) are then used to compute the relative
magnitude errors in dB. The linear components of the coherent Figure 5. Coherent Phase Error versus Error Ratio
error are then computed with (3) and (4). For example, if the
signal to coherent error ratio is 40 dB, the linear error is 0.01. D. Example Coherent Factor
Figure 4 shows the well-known plot for error in dB of a coherent Stray signal error is a primary coherent error. The stray
error [4]. signal error is usually expressed as so many dB down from the
signal level arriving at the antenna under test along the range
axis. This stray signal level is always there. When the AUT
peak directivity in the pattern is aligned to the range axis, that
signal will have a stray signal level that is that many dB down
from the measured value. For example, if the range stray signal
level is -55 dB (used in this paper as +55 dB) at a given
frequency, the error of the peak directivity is approximately
±0.001778 dB and the phase error would be approximately
±0.10188 degrees. As the AUT rotates, the antenna response
decreases, bringing the signal measured at that angle closer to
the stray signal level. For a signal 20 dB lower in the pattern,
the stray signal error ratio would be +35 dB, for an amplitude
error of approximately ±0.01778 dB and phase error of
approximately ±1.01893 degrees.

E. Post-Processing Enhancements
Figure 4. Coherent Error Magnitude versus Error Ratio The signal to coherent error ratio can be improved with
certain post-processing techniques. For example, near-field to
C. Calculation of Coherent Error Phase far-field transformations will further discriminate against
While the coherent error equations are well known, the phase coherent errors, with spherical near-field processing providing
error due to a coherent signal is less so. In [7], a method of the most improvement. Advanced Antenna Pattern Comparison
calculating phase error was presented, which was revised by [8] (AAPC) [9] can also improve the stray signal ratio. Spatial
filtering techniques [10][11] are designed to provide stray signal B. Calculation of the Non-Coherent Error Magnitude
suppression. Time domain gating [12] is also designed for stray Equation (6) is applied to the non-coherent signal to error
signal suppression. The value of any post-processing stray ratio and then equations (9) and (10) are used. Equations (3) and
signal suppression should be added to the base coherent signal (4) are then used to obtain the relative magnitude errors in dB.
to error ratio before applying (6). For example, if the stray signal
ratio is +50 dB and AAPC techniques are used, the stray signal Note that for equality of the non-coherent error and the
level improves to +65 dB or so. desired signal, the peak positive error is now 3 dB. Figure 7
shows the non-coherent error curve versus signal to error ratio.
VI. NON-COHERENT CALCULATIONS
All error signals that do not meet the criteria to be coherent
are non-coherent. The primary example is the ratio of the
desired signal to the thermal noise level of the receiver in the test
system.

A. Model
Reference [7] addressed the statistical nature of noise by
defining a “peak noise level”. This level is the number of dB
above the receiver noise floor at which the evaluator wants to
treat as 0 dB S/N. Within this range, the evaluator can expect to
have all noise values to a high statistical probability. For
example, assuming the noise is Gaussian, selecting a peak noise
level of 8 dB above system noise will have a probability Figure 7. Non-Coherent Error Magnitude versus Error
(confidence level) of 0.993 of containing all possible noise Ratio
values. This is approximately a 3 σ criterion. This gives the
noise vector a circular boundary like the coherent error
boundary. This is shown in Figure 6, which is the non-coherent C. Calculation of Non-Coherent Error Phase
version of Figure 3. The calculations for the non-coherent phase error are the
same as for the coherent case using equations (12a) through
(12d). Therefore, Figure 5 is also a plot of the non-coherent
phase error versus the error ratio.

D. Example Non-Coherent Factor

Thermal noise error is a primary non-coherent error. The
receiver or network analyzer used for antenna measurements
usually has an adjustable noise floor, set either by a sample rate
or an IF bandwidth. The peak signal in the pattern is related to
the noise floor based on the receiver reading. For example,
suppose a receiving system experiences 0.1 dB compression at -
15 dBm and its adjustable noise floor is set to -100 dBm. If the
a) b) c) peak noise ratio is chosen to be 8 dB, then the peak signal noise
floor is -92 dBm, giving 77 dB of dynamic range. For most
receivers and VNAs, the 0.1 dB compression level will be the 0
Figure 6. Non-Coherent Errors a) Error < Signal dB reading on the scaled output. Therefore, if the receiver reads
b) Error = Signal c) Error > Signal -12 dB, that is an absolute level of -27 dBm, placing the peak
pattern signal at 65 dB above the peak noise signal. This results
The difference is that while the coherent error will always be in a ±0.0000014 dB error. Other points in the pattern will have
on the circle, the non-coherent error can be anywhere inside the correspondingly lower signal to error ratios.
circle. For the situation of Figure 4 a), the peak error level
provides a worst case amplitude error for the confidence level E. Post-Processing Enhancements
chosen. As we reach the equality case of Figure 4 b), there is a
possibility of a noise vector having equal magnitude and Non-coherent errors can have their effects reduced by
opposite phase of the desired signal, giving a 0 resultant. As with providing increased raw dynamic range during the
the coherent case, the phase error approaches ±90 degrees [8]. measurement, at the cost of data collection throughput. Post-
For the case of the desired signal being smaller than the peak processing improvements can be obtained by any technique that
noise boundary, note that the equality case is included in the area includes DFT/FFT processing of the data. These include the
of the circle. Therefore, worst case negative error remains at a near-field to far field transforms [13]. The DFT/FFT processing
level of -∞ dB for any time the peak noise level is greater than in these techniques includes the well-known noise suppression
the desired signal. Note also that the resultant phase can be any of:
value once the non-coherent error is larger than the desired #
𝑆𝑢𝑝𝑝𝑟𝑒𝑠𝑠𝑖𝑜𝑛 = 10 ∗ log (13)
signal, as with the coherent phase error.
 For NF-FF processing, the noise suppression should be Parameter System or Derived Values
subtracted from the native 0 dB noise level of the receiver. Measurement
Values
Mechanical Alignment +/-0.12 dB
VII. EXAMPLE DYNAMIC RANGE ERROR ANALYSIS magnitude error in dB
The example dynamic range error analysis consists of stray Mechanical Alignment +/-0.6 deg
signal (coherent) error, thermal noise (non-coherent) error, phase error in degrees
receiver linearity error (coherent with receiver reading) and
mechanical alignment (constant) error. For this study, the errors The suggested process for creating a spreadsheet to perform
were computed over the receiver dynamic range for the data, the analysis is as follows:
which would be from the receiver compression point down to
the 0 dB receiver noise floor. Figure 8 shows the significant  On the first worksheet, enter the data from Table I and
signal levels used in the analysis, while Table I shows the make the computations indicated in the table.
parameters and data needed to perform the example analysis.
 Create a worksheet for each error and compute the +/-
Receiver/VNA Compression Point dB and +/- linear error terms along with the phase error
terms for that error. Each row in the worksheet should
Maximum Signal in the Pattern be a dB increment in the receiver dynamic range,
indexed to the peak signal in the pattern.
 Create a worksheet to compute the composite RSS
magnitude errors versus dynamic range, taking the
Receiver Dynamic Range
individual linear errors from the individual error
worksheets.
Stray Signal Level Based on Max Signal  Create another worksheet to compute the composite
RSS phase errors versus dynamic range.
Selected Peak Noise Level
0 dB Noise Level  It is helpful to have plots of the results on each error tab
and the composite tabs. All plots in this paper were
Figure 8. Important Signal Levels for the Analysis generated in the authors’ prototype spreadsheet.
The user can validate their implementation of the analysis
TABLE I. DATA AND DERIVED VALUES FOR THE ANALYSIS. spreadsheet by using the example values and verifying that the
spreadsheet results match the plots in this paper. Figure 9 shows
Parameter System or Derived Values the composite amplitude error versus pattern dynamic range.
Measurement
Values
Selected peak noise level 8 dB
ratio
Receiver 0.1 dB -15 dBm
compression point
Receiver reading at the 0 dB
compression point
Selected receiver noise floor -100 dBm Receiver dynamic range
= 85 dB
Peak noise level = -92
dBm
Estimated/Measured stray -55 dB Used as +55 dB
signal level
Receiver amplitude linearity +/-0.04
dB/decade
Receiver phase linearity +/-0.5°/decade
Post-Processing stray signal 0 dB
improvement
Post-Processing thermal 0 dB Figure 9. Composite Amplitude Error
noise improvement
Peak signal absolute Note that the positive error is a smoothly increasing curve.
level = -27 dBm The negative error curve shows the interaction of the stray signal
Absolute Stray signal and thermal noise negative curves at the -∞ dB points for each.
level = -82 dBm The RSS process smooths this interaction in that the spreadsheet
Peak signal is in the 2nd
Pattern Peak Signal in the
-12 dB decade of the receiver
was set up to limit the minimum negative error to -100 dB. Note
receiver scale that these curves represent the absolute worst case error bounds.
dynamic range
Effective dynamic range
= 73 dB
Figure 10 shows the composite phase error. Note that once
Peak signal S/N ratio = the stray signal error passes ±90 degrees, the error quickly goes
65 dB to the ±180 degrees limits.
 given antenna pattern and contains the individual error
contributors at that signal level. Additional observations:
 The stray signal error or thermal noise error will
dominate, depending on which one occurs first as the
pattern signal moves lower in the dynamic range.
 Post-processing techniques to improve stray signal or
thermal noise errors can improve the error boundaries
considerably.

REFERENCES
[1] P149™/D5 “Draft Recommended Practice for Antenna Measurements”,
IEEE, May 2019, unpublished.
[2] IEEE Std 1702-2012, “IEEE Recommended Procedures for Near-Field
Antenna Measurements”, IEEE, 2012.
Figure 10. Composite Phase Error [3] P1502/D1, “Draft Recommended Practice for Radar Cross-Section Test
Procedures”, IEEE, October 2016.
The user can employ the spreadsheet results to determine the
error bounds by selecting a point on the pattern, determining the [4] Searcy Hollis, Jud Lyon, Larry Clayton, Antenna Measurements, Chapter
Chapter 14.2.4, pp. 14-25-26, Scientific-Atlanta, 3rd edition, 1985.
number of dB down from the pattern peak and looking at that
[5] Stephen Blalock and Jeffrey Fordham, “Estimating Measurement
level in the spreadsheet. An alternative method is to import the Uncertainties in Compact Range Antenna Measurements”, AMTA
error bounds into a data presentation function and show the error Annual Symposium, 2015.
bounds graphically, as shown in Figure 11 for a segment of an [6] NIST Technical Note 1297, “Guidelines for evaluating and expressing the
antenna pattern. Uncertainty of NIST Measurement Results”, 1994 edition.
[7] John Swanstrom, “Measurement Considerations for Antenna Pattern
Accuracy”, AMTA Annual Symposium, 1997.
[8] Ryan Cutshall and Jason Jerauld, “Revising the Relationships between
Phase Error and Signal-to-Noise Ratio”, AMTA Annual Symposium,
2015.
[9] Celestino Corral, “Practical Issues in Advanced Antenna Pattern
Comparison”, AMTA Annual Symposium, 1996.
[10] Doren Hess and Scott McBride, “Evaluation of Isofilter™ Fidelity in
Selected Applications”, AMTA Annual Symposium, 2008.
[11] Greg Hindman and Allen C. Newell, “Mathematical Absorber Reflection
Suppression (MARS) for Anechoic Chamber Evaluation &
Improvement”, AMTA Annual Symposium, 2008.
[12] Andrew Predoehl and Warren Stutzman, “Implementation and Results of
a Time-Domain Gating System for a Far-Field Range”, AMTA Annual
Symposium, 1997.
[13] J. E. Hansen, Spherical Near-Field Antenna Measurements, page 371,
IEEE (Peter Peregrinus, Ltd.), 1988.
Figure 11. Portion of an Antenna Pattern with Error Bounds

VIII. SUMMARY
This paper has consolidated numerical computation of
amplitude and phase errors for four categories of antenna pattern
errors. A methodology to combine these errors versus the
dynamic range of an antenna pattern has been presented that
supports determining the error bounds for any signal level in a