Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
I. Introduction
Development in the direction of digital systems means that new methods of measuring
interference sources must be developed. The present emission requirements have been
developed to protect analog radio. These limits are defined as the maximum allowed level of
the measured quasi-peak value of the radiated emission from the interference source.
However, the levels measured by the quasi-peak detector are not correlated to the impact of
an interference source on a digital radio system. The quasi-peak detector was originally
developed to simulate the human perception of electromagnetic interferences on analog radio
receivers. Furthermore, the limits are only defined for the frequency band below 1 GHz. As
several radio services already operate beyond 1 GHz, there is a great need for new
requirements [5, 20].
The Amplitude Probability Distribution (APD) has been discussed as a possible measure of
the radiated interference that would indicate the degradation of a digital radio receiver [32].
The APD was used in the late 1960s and 1970s mainly to characterize interference sources
[4,21,22] and in recent years has been discussed conserning its correlation to the bit error
probability (BEP) of digital radio systems. In particular, the relation between the APD and the
impact of microwave ovens on the performance of a certain digital receiver has previously
been presented in [1-3]. However, the literature lacks of a theoretical description of the
connection between the APD and the performance of digital communication systems. The
correlation has mainly been demonstrated by measurement, but in [1] a theoretical relation
between a microwave oven and a certain receiver is shown. As the same paper assumes no
AWGN, the approach requires several new expressions for the BEP. Depending on the energy
of the contribution from the interference signal, different error expressions are required.
Furthermore, error expressions for different receivers need to be derived to analyze the
performance of a communication system with an arbitrary modulation method. The need for a
theoretical description of the impact of a digital receiver and the APD of an interference
source was also expressed as an issue that remains to be solved in [31].
To evaluate the impact of an interference signal on a digital radio receiver, the interference
signal needs to be characterized in such a way that it can be used in performance estimation.
By tradition, interference signals are often modeled as Gaussian processes. Unfortunately, an
approximation of impulsive noise or pulse-modulated noise as additive white Gaussian noise
often results in an underestimation of the resulting BEP of a victim receiver [33]. Hence,
especially for these two types of interference, it is of great interest to consider the statistical
character of the interference signal. Several papers model interference signals as impulsive or
non-Gaussian noise. In [6], a coherent Binary Shift Keying (BPSK) receiver subjected to an
interference signal with arbitrary amplitude probability density function (pdf) is considered;
whereas the error probability for some receivers subjected to a class A interference (an
interference model for which the bandwidth is narrower than the radio receiver of interest) is
examined in [7-9]. The performance of impulsive interference in single-user systems and
multi-user systems, respectively, has been studied in [10-11]. Furthermore, the problem of a
communication system subjected to a non-Gaussian interference environment is treated in
[12-14]. However, in these papers the digital radio receivers are assumed to be sub-optimized
to Gaussian noise and robust against deviations from Gaussian noise. For the special case with
a receiver optimized for AWGN, these papers provide error expressions of this receiver
degraded by non-Gaussian noise. However, these performance expressions are complicated
and their usage in practical applications considering a Gaussian optimized receiver is not
obvious.
The paper is partly based on results published by the author in [23-24]. The paper provides a
systematic and practical method of incorporating the APD measure in conventional error
expressions developed for AWGN. The method also makes it possible to consider the impact
of an arbitrary interference source in the general error expressions originally derived for
AWGN. Furthermore, the relation between the maximum error probability of a digital
receiver and the measured APD of an interference source is stated. This relation opens the
possibility to derive emission requirements for interference sources based on the APD. The
relation between the maximum BEP and the APD is supplemented with an illustration of its
use for emission requirements and has not been published before. However, to estimate the
performance degradation of a digital receiver or to derive emission requirements by the use of
APD, the bandwidth of the measurement receiver and the radio receiver must be similar.
When the bandwidths differ, the APD measured cannot be used directly in these applications.
In [25], a method of converting the APD measured by one bandwidth to another is presented.
The method has been developed for a special group of signals, namely pulse-modulated noise,
which in many scenarios is a relevant type of interference. Reasons for choosing these
interference models are given in [25].
The paper is organized as follows. Section II presents a system model and describes the
problem. In Section III, a measured APD of an interference source is connected to the
degradation of a digital radio receiver due to the same interference. This is presented as a
systematic method which incorporates the use of the APD measure in conventional error
expressions developed for AWGN. The method suggested paves the way for estimating the
performance of digital communication systems in complex interference environments. The
applicability of the method is then shown by an example. Section IV shows that the maximum
bit error probability (BEP) for a BPSK receiver is equal to a certain value of the APD. This
property is then generalized for a variety of receivers. The relation obtained can be used to
define maximum emission limits for electrical equipment in terms of APD. By assuring that
the measured APD of a interference source is lower than the proposed requirements, the
impact on the performance of a variety of receivers is guaranteed not to exceed a given
maximum bit error probability. Section V concludes the paper.
This paper discusses the relation between the measured APD of an interference signal and the
performance degradation on a digital radio receiver due to the same interference, see Fig. 1.
Electrical equipment, such as micro-wave ovens, from which the radiated interference might
have a non-Gaussian amplitude character, has been shown to severely affect the performance
of radio receivers. The key issue is to analyze the information the measured APD provides
and connect it to parameters which are important when the performance of a receiver is to be
estimated. The reverse problem is also of interest, i.e. to relate a certain level of the
performance measure BEP of the radio receiver to requirements on the APD of an
interference signal.
Measurement system
Electrical
equipment
Filter H(f) APD detector APD
1 2
Filter
∫ dt Performance measure, BEP
Radio system
Figure 1: Overview of the problem; connect the measured APD of an interference source to
the performance degradation of a digital coherent radio receiver (1) and the reverse problem
(2).
C. Interference signal
The radio system considered in this work is subjected to an interference environment, which
will negatively influence the performance of the radio receiver of the intended signal. The
interference sources, which constitute the interference environment, are assumed to be co-
located with the receiver. The short distance to the receiver implies that electrical equipment,
even with a moderate level of emission, can constitute a severe problem.
The APD is defined as the part of time the measured envelope of an interfering signal exceeds
a certain level [1]. We assume that the measured signal is ergodic and that the measurement of
the APD is long enough to capture the signal properties. The relation between the APD R (r )
and the probability density function of the envelope, R, is
APD R (r ) = 1 − FR (r ) (1)
and
d d
f R (r ) = FR (r ) = − APD R (r ) , (2)
dr dr
where FR (r ) and f R (r ) denote the cumulative distribution function (cdf) and probability
density function (pdf), respectively.
An APD detector can be implemented by an envelope detector and a counter [16-18]. The
APD can be estimated by a spectrum analyzer, where the signal is first converted to an
intermediate frequency and band limited by a variable resolution bandwidth filter. The signal
can then be compressed by a log amplifier, after which the envelope is extracted by an
envelope detector [16].
To be able to use the information provided by the APD in the following analyses, some
assumptions are necessary. The receiver structures of the APD detector and the analyzed radio
receiver have to be quite similar. This is normally the case for coherent digital radio receivers.
APD gives information about the envelope statistics from the IF filter, which corresponds to
the required information for performance evaluation at the radio receiver. However, the
bandwidths of the measurement and the radio receiver need to be approximately the same. If
the bandwidths of the radio system and the measurement receiver differ significantly, a
method of modifying the APD is suggested in [25] for pulse-modulated interference.
Furthermore, the APD needs to be measured at the frequency band the radio system works on.
The traditional way of determining the error probability of a digital radio receiver is to assume
that the interference can be modeled as AWGN. For that kind of noise, error probability
expressions are often quite easily derived for different kinds of receivers. For other types of
noise, there are no simple methods. But, as will be shown here, with information provided by
an APD detector, even noise of a non-Gaussian nature can be incorporated with the
conventional error probability expressions. The method is then demonstrated in an example.
The key issue when determining the impact of an interfering signal on a coherent digital radio
receiver is information about the envelope and phase at the decision moment in the detector.
For example, if we assume “+1” was transmitted, the decision variable of a coherent BPSK
receiver in AWGN can be described as
y = Eb + n , (3)
where E b is the bit energy and n represents the additive Gaussian noise component, which
has zero mean and variance σ 2 = N 0 2 . Thus, the performance is obtained as [27]
⎛ Eb ⎞
Pb = Q⎜⎜ ⎟,
⎟ (4)
⎝ N 0 2 ⎠
where
∞
1 ⎛ x2 ⎞
Q(v ) = ∫ ⎜⎜⎝ − 2 ⎟⎟⎠ dx .
exp (5)
2π v
For a BPSK receiver subjected to an interfering signal, the decision variable Y has the
conditional expected value E [Y r , ϕ ] = E b + r cos ϕ and the variance σ 2 = N 0 2 , where
r cos ϕ is the contribution from the interference. In detail, r and ϕ denote the envelope and
the phase, respectively, of the interference. Thus, the conditional error probability, adjusted
for the interfering signal, becomes
⎛ E b + r cos ϕ ⎞
Pr [bit error r , ϕ ] = Q⎜ ⎟ . (6)
⎜ N 2 ⎟
⎝ 0 ⎠
By assuming that the phase in the moment of decision is uniformly distributed over [0,2 π ]
and by using the information from the APD, the error probability is obtained as
2π ∞ ⎛
1 E b + r cos ϕ ⎞
⎜ ⎟ f R (r ) dr dϕ .
2π ∫0 ∫0 ⎜⎝
Pb = Q (7)
N0 2 ⎟
⎠
At this stage, the information from the measured APD can be used to provide the probability
density function of the envelope f R (r ) . This is based on certain assumptions, e.g. that the
bandwidth of the measuring detector is approximately the same as that of the analyzed radio
receiver, see section IID. By modifying E b with E b + r cos ϕ in the conventional error
expression and then averaging over the envelope and phase, the influence of the interference
is considered. This method can be generalized to other coherent digital modulation schemes
with bit-by-bit decisions. The method includes Gaussian noise, originated from thermal noise
in the receiver. If only the interference is to be considered, which means that the thermal
receiver noise is neglected, N 0 can be made arbitrarily small in practice. The method can be
summarized into the following steps:
B. Example
The symbol error probability of a Quadriphase-Shift Keying (QPSK) modulated signal can be
derived with the same approach used in section IIIA. In order to evaluate the influence on a
two-dimensional modulation scheme such as coherent QPSK, the contribution from the
interfering signal also has to be described in two dimensions. To demonstrate the method, we
assume an interfering source that emits pulse-modulated Gaussian noise. Measurement
equipment with an APD detector measures the interfering signal. The measured pulses have a
pulse width Tp , which come periodically with a period time of T . This gives a duty factor of
α = Tp T .
The pulses and the noise between the pulses are characterized by Gaussian distributed noise
with the variance σ 1 and σ 2 , respectively. Nevertheless, the final pdf exhibits a non-
2 2
This model often suits well as a model for signals radiated from electrical equipment [25].
The APD has been calculated for this interfering signal with the current parameters
σ 12 = 10 , σ 2 2 = 1 and α = 0.1 and is shown in Fig. 2. The APD
0
10
−1
10
−2
10
−3
10
APD(r)
−4
10
−5
10
−6
10
0 1
10 10
Noise envelope, R
does not indicate the order in which the envelope samples come in time. If the samples are
switched in time, they are still characterized by the same APD. It is worth noticing that as
long as the detector takes bit-by-bit decisions, which are used for signals without memory,
this does not matter. Only the statistics of r and ϕ are of importance for the performance.
This implies that you can create an APD through a polynomial or a simple mathematical
model equal to a measured APD of a microwave oven, for example, and use the simpler
model when the impact is to be determined. With the previously described parameters and
E b = 100 , the bit error probability can be calculated, see Fig. 3. The figure also shows the bit
error probability in the absence of interference source when only thermal receiver noise is
present.
0
10
−1
10
Bit error probability, BEP
−2
Disturbance consisting
10 pulse modulated noise
−3
10
−4
10
No disturbance
−5
10
−4 −2 0 2 4 6 8 10 12 14
Signal−to−noise ratio, Eb/N0, [dB]
Figure 3: Estimated bit error probability of a QPSK signal with and without interfering
pulse-modulated Gaussian noise.
As QPSK symbols are mapped by two information bits, the symbol energy Es = 2E b , where
E b denotes the bit energy. Here the contribution from the interfering source to the decision
variable is defined with Es 2 + r cos ϕ instead of E s 2 in the inphase channel and with
Es 2 + r sin ϕ instead of Es 2 in the quadrature channel. By substituting Es 2 with
Es 2 + r cos ϕ in the inphase channel and Es 2 with Es 2 + r sin ϕ in the quadrature
channel, the conditional symbol error probability is obtained as [27]
⎛ Es 2 + r cos ϕ ⎞
Pr (symbol error r , ϕ ) = Q⎜ ⎟
⎜ N0 2 ⎟
⎝ ⎠
⎛ Es 2 + r sin ϕ ⎞ ⎛ Es 2 + r cos ϕ ⎞
+ Q⎜ ⎟ − Q⎜ ⎟. (9)
⎜ N 2 ⎟ ⎜ N 2 ⎟
⎝ 0 ⎠ ⎝ 0 ⎠
⎛ Es 2 + r sin ϕ ⎞
* Q⎜ ⎟
⎜ N 2 ⎟
⎝ 0 ⎠
Finally, the expression is averaged over r and ϕ . By assuming that the phase is uniformly
distributed over [0,2 π ] at the moment of decision, by using the symmetry of the cosine and
sine function and by assuming that the last term is relatively small, the symbol error
probability can be written as
Pr (symbol error )
1
2π ∞ ⎛ Es 2 + r cos ϕ ⎞ . (10)
⎜ ⎟ f R (r ) dr dϕ
≤
π ∫0 ∫0 ⎜
Q
N0 2 ⎟
⎝ ⎠
Furthermore, by using the assumptions of Gray coded symbols, the bit error probability can
be approximated as [28]
1
Pb = Pr (symbol error ) . (11)
2
By assuming statistical independence of the noise quadrature carriers, the relation between the
maximum BEP and the APD can be generalized for other coherent modulation schemes by
studying the signal constellation. For signals that exhibit statistically dependent quadrature
components, it has been shown that the resulting BEP of a Quadrature Amplitude Modulation
(QAM) system is only marginally affected by this property [29]. This means that the proposed
relation might be useful in practice also for situations when the quadrature components are
statistically dependent.
The distance between the closest symbols is defined as the minimum distance and is of
significance for the error probability of a coherent detector. The worst case symbol error
probability is achieved when the contribution from the interfering signal is directed toward the
closest symbol in the signal constellation. Considering the worst case, a symbol error occurs
when the envelope of the interference exceeds d min 2 . This is due to the fact that the border
between decision regions is, in a conventional coherent receiver, located in the middle
between two symbols. Therefore, the symbol error probability conditioned on a worst case
phase value Pr[symbol error] max can be obtained as
⎡ d ⎤ ⎛d ⎞
Pr[symbol error ] max = Pr ⎢r > min ⎥ = APD R ⎜ min ⎟ . (15)
⎣ 2 ⎦ ⎝ 2 ⎠
The expression shows that the symbol error probability is always less than or equal to the
APD for a certain value. This implies that the APD of a measured interference source for the
value equal to d min 2 must not exceed the maximum acceptable bit error rate. By letting the
measured APD for the value d min 2 be less than the determined maximum allowed error rate,
the error rate will always be lower than or equal to what is acceptable. Such reasoning makes
it possible to define requirements on the APD based on requirements on a BEP level.
( )
Eq. (15) can be rewritten as Pr[symbol error ] wc = APD R β E b , where β takes different
values depending on the modulation scheme. The minimum distance for an M-ary Phase Shift
Keying (PSK) signal is [27]:
⎛ ⎛ 2π ⎞⎞
d min = 2 log 2 (M ) E b ⎜⎜1 − cos ⎜ ⎟ ⎟⎟ , (16)
⎝ ⎝M ⎠⎠
For example, the bound of the BEP for a coherent BPSK receiver subjected to an interference
can be interpreted as follows. If the measured bit energy at the detector is E b ,1 , the maximum
( )
BEP never becomes higher than Pb,1 = APD E b,1 , whereas for a smaller bit energy E b,2 the
( )
BEP is bounded by the larger value Pb,2 = APD E b,2 , see Fig. 4.
APDR(r)
Pb,2
Pb,1
To demonstrate how the derived bounds on the BEP can be used for emission requirement, we
assume that the BEP is restricted never to become higher than 1 ⋅ 10 −3 , which corresponds to a
typical requirement for voice transmission. For example, the subjective effect of bit errors for
voice transmission with pulse code modulation (PCM) is: 1 ⋅ 10 −6 not perceptible; 1 ⋅ 10 −5
single clicks; 1 ⋅ 10 −4 single but little distracting clicks; 1 ⋅ 10 −3 high density of clicks, which
disturb each speech level; 1 ⋅ 10 −2 strong disturbing crackle with low intelligibility [30].
Furthermore, the value of the bit energy at the detector must be determined. Here, we assume
that E b = 10µV . Inserting the determined values of the maximum allowed BEP and the bit
energy in the bounds described in Table 1, the requirements can be implemented as a point in
the APD for every modulation scheme. To ensure that the error rate of a receiver that is
subjected to an interfering signal does not exceed a certain error rate, the measured APD must
lay below the points, as illustrated in Fig 5. The figure shows measured data reported in [1]
concerning radiated interference from two different microwave ovens: A: Inverter-type of 600
W at 2.45 GHz and E: transformer-type of 500 W at 2.45 GHz. In the same figure, the
requirements, corresponding to a maximum BEP of 1 ⋅ 10 −3 and E b = 10µV , are displayed
with circles. If the measured APD lays below all the points, the impact on different radio
systems in Table 1 are guaranteed not to exceed the permitted level of the BEP. We can see
that the microwave oven of transformer-type (E) fulfills the requirement both at 1 m and 3 m.
This means that the radiated interference of this microwave oven will not cause a bit error rate
worse than 1 ⋅ 10 −3 . However, the micro-wave oven of inverter-type (A) does not fulfil the
requirement, and thus we cannot guarantee that the bit error rate is lower than the
requirement, although the average BEP might be lower than the requirement. It is important to
note that the requirement only places restrictions on the APD levels in a specified noise
envelope interval. In the example shown in Fig. 5, the APD is restricted between 3.8 ⋅ 10 −6 V
and 10 ⋅ 10 −6 V. For a higher or lower noise envelope, the interference can assume arbitrary
APD levels.
−1
10
−2
10
−3
10
APD (Prob[e>E])
−4
10
−5
10
Measured data, MWO A, 1 m
Measured data, MWO E, 1 m
Measured data, MWO A, 3 m
Measured data, MWO E, 3 m
−6 Requirement acc. to Table 1
10
−6 −5 −4 −3
10 10 10 10
Noise envelope, E, [V]
Figure 5: Illustration of the bounds implemented in an APD graph with measured
radiated interference from two different microwave ovens at 1 m and 3 m [1]. The
requirements are calculated for a maximum BEP of 1 ⋅ 10 −3 when E b = 10µV .
IV. Conclusions
It has previously been stated that the APD of an interference signal is strongly correlated to
the BEP of digital radio receivers. However, there the literature lacks of a theoretical
description of the relation between the APD of an interfering signal and the impact on a
digital receiver. This paper summarizes [23,24] with the aim of clarifying the theoretical
relation between the APD and the BEP. This paper presents a method of how to use these
results in practical applications. It demonstrates that the APD provides the necessary
information about an interference signal to estimate its degradation of a digital coherent radio
receiver under certain conditions. Estimation of the impact of an arbitrary interference on
digital coherent receivers has been presented in [12-14]. However, analyzing the performance
of a receiver optimized to AWGN that is subjected to non-Gaussian interference constitutes a
special case. These expressions are complicated and their use in practical applications is not
obvious. This paper proposes a systematic method of incorporating the contribution of an
interfering signal, which might be non-Gaussian provided by an APD measurement system, in
conventional error expressions developed for AWGN. The paper also suggests a possible
approach to defining emission requirements on the APD in order to control radiated
electromagnetic emission for the protection of radio communication systems.
References