309
to find values of the coefficients
51
...zz
in channeling
equation (1) from the given image of response in the
DFD. Having these values, the user can determine the
most probable physical effects which cause the emer-
gence of the nonlinear interference corresponding to the
recognized response.
Analysis of Response Recognition Methods
Known methods of response recognition are con-
sidered in [1, 2]. Let us analyze their drawbacks:
1) Method of slope angle [1, Section IV, method 1]:
1.1) Capabilities of the method are limited: without
additional measurements, it is possible to determine only
the coefficients
1
z
and
2
z
; the coefficient held by the
frequency of one of the local oscillators (as a rule,
3
z
)
can be estimated by changing the tuning frequency of the
receiver under test (this requires to repeat the measure-
ment of the double-frequency characteristic once again);
the coefficients held by the frequencies of the remaining
local oscillators can not be determined by the method.
1.2) The method accounts for the slope angle of the
response image, but it does not take into account the
position (displacement) of the image – this may cause
errors in recognition. For example, even-order inter-
modulation responses of the kind
)0,,(),,(
321
mmzzz −=
,
...,2,1
m
have equal slope angles
45
[1,
Fig.6(e)], therefore the method is unable to distinguish
such responses from each other.
2)
Method of patterns’ (specific groups of response
images) recognition [1, Section IV, method 2]:
2.1)
The method is not always able to solve the ba-
sic problem – to recognize a single image of response
(one arbitrary line in DFD). Let us give examples of
problematical situations: a) most of the specific groups
are recognized by their appearance at the point of inter-
section of component parts (response images), but that
point may be located outside of the DFD’s measurement
area
)},(),,{(
max2min2max1min1
ffff
; b) some response
images being parts of a specific group may be fully or
partially invisible (this may cause wrong recognition of
the group’s order).
2.2)
For receivers having more than one frequency
conversion, the position of several specific groups
(nodes of order higher than one) may be unobvious – this
complicates their recognition and may cause errors.
That is why the method can be used as an additional
method only, although it is very convenient for heuristic
analysis of DFD by human-operator.
3)
Method of frequency measurement and solving
the system of linear algebraic equations (SLAE) [1, Sec-
tion IV, method 3]:
3.1)
The number of equations in the system is rig-
orously defined – it is equal to the number of unknown
coefficients
51
...zz
, so we get
2+
fc
N
equations –
ref. (1) and (2). This limits the recognition accuracy by
the errors of one-shot measurements.
3.2)
The system of equations is solved in real num-
bers and the obtained values of coefficients
51
...zz
are
rounded to the nearest integers, but such approach does
not guarantee that the best integer solution will be
achieved.
3.3)
Additional equipment is needed for measure-
ment of frequencies of local oscillators, test signals, and
receiver-under-test output signal.
4)
Method of modulation parameters’ comparison
between the input test signals and the output signal [1,
Section IV, method 4]:
4.1)
Limited capabilities (ref. item 1.1).
4.2)
Increased requirements to the measuring gen-
erators because the test signals must have user-defined
parameters of modulation.
4.3)
Additional equipment is needed for measure-
ment of the output signal’s modulation parameters.
5)
Method of frequency change rate comparison be-
tween the fast-changed test signal and the output signal
[1, Section IV, method 5]:
5.1)
Limited capabilities (ref. item 1.1).
5.2)
Additional equipment is needed for measure-
ment of the output signal’s frequency change rate.
5.3)
For digital frequency sweep, the method can
not be used directly and requires improvement.
5.4)
The method imposes constraint on the test sig-
nal’s frequency change rate for analog sweep (or on the
frequency change step for digital sweep).
As follows from the considered drawbacks, the only
method that is able to solve the recognition problem
completely (i.e., to find all the coefficients
...,,
21
zz
of
an arbitrary response for receiver having any number of
frequency conversions) is “Method of frequency meas-
urement and solving SLAE” [1, Section IV, method 3].
Proposed Recognition Technique
In this paper, we propose a recognition technique
which is obtained by development of the SLAE method
[1, Section IV, method 3] in the following directions:
1) In order to eliminate the drawback given in item
3.1) of Section 2, i.e., to improve the precision of identi-
fication by averaging the results of a large quantity of
measurements, we propose to use an overdetermined
SLAE of the kind (3) and to solve it in the least-squares
sense. For example, as it follows from (1) and (2), we get
the following SLAE if
3=
fc
N
:
==⋅+
+⋅+⋅+⋅+⋅
,...,2,1,
,,35
,24,13,22,11
rpioutiLO
iLOiLOii
Niffz
fzfzfzfz
(3)
where
rp
N
denotes the number of recognition points
);;;;;(
,,3,2,1,2,1 ioutiLOiLOiLOii
ffffff
in which the simul-
taneous measurement of frequencies is performed;
i
is
the index of the recognition point under consideration.
The number of equations in SLAE (3) is arbitrary and it
is equal to the number
rp
N
of recognition points.
2) High performance of modern PCs makes it pos-
sible to remove the drawback given in item 3.2) of Sec-
tion 2 in the simplest way: in this paper, the solution of
SLAE is found by exhaustive search through all potential
solutions (i.e., integer combinations
),,,,(
54321
zzzzz
)
of order not higher than the maximal recognition order
max
M
specified by the user: