Practical
Smart Sensor System
The objective of this experiment is to get introduced to the working principles of different
filter topologies and their respective applications.
It is assumed at the end of the experiment that one should be able to achieve sufficient
knowledge on the structure and operation of the filter. Independent study and learning about
the relevant instructions and additional literature are required to fulfill the objectives of the
experiment and successfully complete the Practical lab.
Literature
Williams, A.;Taylor, F.: Electronic Filter Design Handbook; 4. Edition; Verlag: McGrawHill; 2006
Tietze, U.; Schenk, C.: Halbleiter-Schaltungstechnik; 10. Auage, Springer-Verlag Berlin
Heidelberg,1993
Seifart, M.: Analoge Schaltungen; 4. Auage; Verlag Technik GmbH, Berlin, 1994
Bernstein, H.: Analoge und digitale Filterschaltungen; VDE-Verlag, 1995
Gldner, K.: Systemanalytische Grundlagen der automatischen Steuerung; Gesellschaft
fr Wissens- und Technologietransfer der TU Dresden mbH, 1999
1. Background literature
1.1. Feedback
The principle of feedback: Under the conditions of linearity, with absence of feedback
signal flow applies in one direction:
G
Xa
V
X e 1 VK
(1)
Until | 1 + V K |> 1, there is negative feedback. If the amount is less than 1, it is positive
feedback.
In Equation (1), the reason behind using OPVs very often is: If | V | very large,
1
K
(2)
This means that the behaviour of the overall circuit is solely due to the feedback branch.
The complex transfer function G is frequently illustrated using Bode Diagram, also known as
Bode plot. This diagram shows the amplitude response and the phase response over a
common frequency axis. The amplitude response is obtained from the magnitude of the
Transfer function:
| G( s) | Im2 Re2
(3)
( s) arctan(
Im
)
Re
(4)
U d U P U N U A U D . AD AD
U A
U D
(5)
differential voltage at the input UD 0V is maintained, because the voltage at the output is
phase shifted by 180 degrees to UN. The output voltage thus counteracts its own original
cause. Depending on which input the input signal is applied, one can distinguish between
inverting and non-inverting mode operation.
A( j )
UA
Z
1
UE
Z2
(6)
UA
Z
1 1
UE
Z2
(7)
In a non-inverting amplifier, the gain is always greater than 1. Other important characteristics
of OPVs in practice are: noise performance, offset, CMRR, slew rate and common mode
rejection. For consideration of these properties with respect to the realization of an analog
filter, reference is made at this point to [2].
1.3 Fundamentals of filters
A filter is primarily a frequency selective, that is, the attenuation or the gain change with the
frequency. Such circuits can be realized in various forms, for example, passive filters, active
filters, digital filters, switched-capacitor filter, mechanical filter, CTD-filter, etc. In spite of
the different forms of implementation, all filters can be classified according to type,
characteristics and border frequency.
Filter types are lowpass, highpass, bandpass, bandstop and allpass. The known filter
characteristics are Bessel, Butterworth, Cauer and Tschebyscheff. The mathematical
description of a filter is carried out by the transfer function. This generally has the form
A( P)
Z ( P) 0 1P 2 P 2 ... n P n
N ( P) 0 1P 2 P 2 ... n P n
(8)
0 1P 2 P 2 ... n P n
(9)
(10)
Here the order of the denominator polynomial N(P) indicates the order of the filter and the
coefficients a and b determine the filter characteristics. 0 is a factor that indicates the basic
gain of the filter. Also the gain in a particular frequency range, in which the filter does not
have an influence on the signal. For all further considerations that factor should always be 1.
(11)
( P)
Im[ A( P)]
Re[ A( P)]
(12)
And from the phase transition the total response time can be calculated as:
TGr
d ( j )
d
(13)
The diagram below gives an introduction to the mathematical modeling of filters, based on the
low pass filter. From the voltage divider rule, the transfer function is calculated as:
U E R 1 C 1 j RC
j
(14)
By defining P = j = j(w/wg) mit wg = 1/RC = 2fg; fg : 3dB break frequency, results the
term:
A( P)
1
1 P
(15)
This is the transfer function of a 1st order (-40dB / dec) low-pass filter (TP). For comparison,
the overall transfer function of such a filter:
A( P)
A0
1 1 P
(16)
2
1
U E R j L
C 1 j RC ( j ) LC
j
1
A( P)
1 g RCP g 2 LCP 2
(17)
A0
1 1 P b1P 2
(18)
(19)
In order to realize a higher-order filter, you can cascade several 1st and 2nd order filters
together. However, it should be noted that the cutoff frequency fges of the cascaded filter is
not equal to the cut-off frequency of the individual filters.
f ges f x
(20)
0
1
1
1
1
(1 1 1 2 )(1 2 2 2 )...
P
P
P
P
(21)
A( j )
R
1
R
jC
A( P)
1
1
1
j RC
(22)
(23)
1
1
P
1
1
1
1
(P )
P
1 P P 2
(24)
Although you will find a 2nd-order filter, the band pass has still only an attenuation of 20 dB /
dec.
(P) 2
b1
A( P)
a1
() 2 2 a1
1 P [2
]P P3 P 4
b1
b1
b1
(25)
When dimensioning, make sure that the coefficient is derived from LP of 2nd order, although
this is a band-pass of 4th order.
There are two ways to realize a band pass. The first way is to cascade a low pass and a high
pass. The cut-off frequencies of the partial filter must not be selected accordingly and
individually dimensioned. However, this means a double circuit complexity for the filter and
it is particularly suitable for broadband filters. The second way is using resonant circuits and
oscillators. The particular resonance frequency will be selective damped or amplified
(bandpass) thereby.
Wobei die jeweilige Resonanzfrequenz wahlweise verstrkt (Bandpass) ober gedmpft
(Bandsperre) wird. Mit dieser Variante knnen gute schmalbandige Filter aufgebaut werden,
also Filter hoher Gte.
The quality Q is defined as:
fr
r
1
B g ,max g ,min
(26)
B is the 3dB bandwidth, and wr is the resonance frequency of the filter or the center
frequency.
To go from the transfer function of the LP to a band top we replace P by /(P+1/P). A 2nd
order band-stop filter therefore has the form:
A( P)
1
1 P2
1 P P
1
1
P
P
(27)
Basic gain, cut-off frequency and filter characteristics cannot be adjusted separately.
The phase of the signal is shifted by the inverting operation by 180 degrees.
a) Sallen-Key
This is the positive feedback variant. Such 2nd order LP looks like Figure 12 as shown.
R2 and C1 form a 1st order LP, also the gain of the system using C2 will be reduced with
increasing frequency. R3 and R4 must be set properly for the gain, because a positive
feedback has always a risk of instability in the circuit.
R3 (k 1) R4
(28)
A( P)
k
1 Pg [ R1C1 R2C1 (1 k ) R1C2 ] P 2g 2 R1R2C1C2
(29)
A( P)
k
1 2 Pg RC1 P 2g 2 R 2C1C2
(30)
C1
C2
a1
4 f g R
(31)
b1
4 f g Ra1
(32)
Case 2: R1 = R2 = R and C1 = C2 = C,
A( P)
k
1 Pg RC (3 k ) P 2g 2 R 2C 2
(33)
b1
2 f g
(34)
RC
k V0 3
a1
b1
(35)
R2
R1
A( P)
RR
1 g C1 ( R2 R3 2 3 ) P g 2 R2 R3C1C2 P 2
R1
R2
(36)
(37)
R3
R2
A0
b1
4 f g 2C1C2 R2
2
C2 4b1 (1 A0 )
C1
a12
(38)
(39)
(40)
The values of the capacitors C1 and C2 are to be set such that the ratio is not much higher
than the above criterion.
(41)
A( P)
k
R (C C2 ) R1C2 (1 k ) 1
1
1
1 2 1
2
R1 R2C1C2g
P R1R2C1C2g P 2
(42)
With k = 1 and C1 = C2 = C:
R1
R2
1
f g Ca1
(43)
a1
4 f g Cb1
(44)
A( P)
1
2C2 C 1
1
1
2 2
R1CC2g s R1R2CC2g s
C
C2
(46)
2C C2
g R1CC2
(47)
2C C2
g R1CC2
(48)
1 2 A0
2 f g Ca1
(49)
A0
a1
(45)
b1
R1
R2
a1
2 f g C2b1 (1 2 A0 )
(50)
kRCr P
1 RCr (3 k ) P R 2C 2r 2 P 2
fr
(52)
1
2 RC
(53)
k
3 k
(54)
Ar
Q
(51)
1
3 k
(55)
There is obviously a special case, in which all the frequency determining elements, that is, the
resistors and capacitors on the non-inverting input, are dimensioned in each case with the
same values. The low-pass filter characteristic is now solely dependent on the adjusted
internal gain, i.e. the ratio of R1 to R2. With such narrow-band filters, a change of this ratio
alone results in a change in the quality and the gain at the center frequency. This should be
taken into account in the dimensioning. Because if the gain and this signal amplification is the
greater than the quality, then this increase must be either corrected or taken into account in the
analysis of the signal. In addition, it must be ensured that the OPV used, is not brought to its
operating limits, otherwise there will be signal distortions.
g) MFB Band pass filters
A band pass in the MFB variant is shown in Fig. 17. It is simplified by putting C1 = C2 = C:
R2 R3
Cr P
R1 R3
A( P)
2 R1 R3
RR R
1
Cr P 1 2 3 C 2r 2 P 2
R1 R3
R1 R3
fr
1
2 C
R1 R3
R1R2 R3
(56)
(57)
R2
2 R1
1 R2 ( R1 R3 )
f r R2C
2
R1R3
B
fr
1
Q R2C
(58)
(59)
(60)
It can be seen that no parameter can be chosen independently, without having an effect on
other parameters.
h) Band pass filter with Wien-Robinson bridge
Narrow-band filter can also be constructed with the aid of resonant circuits. In this example, a
Wien-Robinson bridge is used.
fr
Ar
1
2 RC
R1 R2
2 R1 R2
R1
2 R1 R2
(61)
(62)
(63)
This circuit is also known as a Wien bridge oscillator in a similar manner. However, as long
as the oscillation condition K * V = 1 is not satisfied, the circuit operates as a band pass.
Theoretically, an infinite quality is reached, however, the circuit then oscillates and is thus no
longer be used as a filter.
i) I-band stop filter
A simple variant, to build a band-stop filter, is the so-called active double-T filter. Generally
it can be developed from each band pass and band stop. Here, the band-pass is included in the
negative feedback of an amplifier. The signal is then attenuated maximum at maximum gain
of the band pass filter.
(64)
k (1 P 2 )
A( P)
1 2(2 k ) P P 2
(65)
fr
1
2 RC
(66)
A0 k
(67)
1
2(2 k )
(68)
Table 1
Filter Characteristics
Z1
Z2 Z3 Z4 Z5
Low pass
High pass
Band pass
In any case, the application of an active filter is a compromise between the improvement of
the signal quality and the influence of the signal. This assessment must be carried out every
time for each application.
A filter is quite often used in metrology to prevent aliasing. In principle it should be
maintained that no frequencies reach higher than half the sampling frequency of the ADC.
However in reality, this is not feasible, because the noise voltages alone bring higher
frequencies with it. Therefore, the maximum amplitude for each application must be
determined and it should not be exceeded. Accordingly, then, a filter must be selected and
dimensioned. As a guideline for this dimensioning, the noise can be used. If the signal is
attenuated so much at half the sampling frequency that it is in the range of noise amplitude, so
there is no benefit with respect to the anti-aliasing feature with further attenuation.
1.6 Noise
Noise is a problem in many measurement applications. When superimposed on the measuring
signal, it leads to the lowering of the detection limit because the measuring signal can only be
distinguished from the noise signal when the amplitudes of the two signals differ sufficiently.
Which amplitude ratio is sufficient for detecting, cannot be defined universally and must be
decided case by case. In order to make this decision, the noise has to be assessed
quantitatively.
Rms noise voltage at thermally rushing resistors:
U rt 4kTBR
(69)
Ideal resistor
Real resistor
(noise)
Noise source
V
B
(70)
ur 2
dU r 2
df
(71)
As already mentioned, the noise voltage has little explanatory power alone. The decisive
factor is the ratio of useful signal to noise signal, the signal-to-noise ratio SNR:
S Ps U s 2
N Pr U r 2
(72)
It should be noted that the noise voltages caused by resistors and OPV, only one part of the
total interference in the signal path. So EMI and mismatches have a much greater influence on
the signal quality than the noise. Reducing these parasitic effects can improve the
measurement signal may more clearly, as lower-noise op amp or smaller resistors.
If there are several sources of noise in a signal path exists, then the noise voltages
geometrically added:
U N U N 12 U N 22 U N 32 ... U N n2
(73)
to available programs. Under certain circumstances this program are sufficient for simple
requirements for order and cut-off frequency. An example of such a program is Filter Free,
the free version to filter Solution of Nuhertz. This program will also be used for the
experiment, since it provides all the necessary functionalities for the experiment already in the
free version. Other notable programs are FilterCad, FiltersCad, FilterPro and FilterLab. To
Operate Filter Free the following notes are helpful. By default, the frequency axis is set to ,
not to f. In addition, the start and end values of the frequency axis must always be entered
manually. You do not change automatically. The key for the test topologies are not referred to
as MFB and Sallen-Key, but as a negative and positive SAB. This term refers to the type of
feedback and the circuit variant because SAB is the abbreviation for Single amplification he
biquad.
The biggest drawback of the program are the fixed resistance values. In practice, it is
customary to determine the capacitance values and adjust the resistance values. These are
usually required for a finer gradation.
This is also generally be observed when using dimensioning of filter programs. The calculated
component values are frequently not available, so there are slight variations of calculated
filter behavior in the physical deployment.
2. Experimental setup
To be able to practically implement theory, the experimental set-up is very important. It offers
the possibility of the tasks to solve by applying the acquired knowledge. Also room for
creativity in solving the tasks set is given. It is desirable to make mistakes, to identify and
resolve.
characteristics of the filter map. To calculate the Bode diagram two measurement signals are
necessary. The input signal of the filter and the output signal. The input signal of the filter is
identical with the output signal of the signal generator. Under ideal conditions, this signal
could be connected to the internal program evaluation. However, since the DA and AD
conversion and the non-ideal properties of the transmission path influence the signal, the input
signal of the filter is measured. This ensures that the difference between the two signals are
caused by the filter.
The test program itself is divided into two levels. A measuring level and an evaluation level.
Only one is visible at one time. The levels are selectable on two riders with appropriate
labeling. This structure was chosen because it simplifies program enhancements. So
additional tabs can be used to insert additional program options or to improve the clarity of
each level.
In the measurement plane both reference signal and the filtered signal are shown in separate
graphs. In addition, amplitude and phase response are also available. At this level, all the
necessary adjustments should be made. The sequence control of the measurement can be
found here. By pressing the "Evaluate" button in the current Bodeplot, if desired, is written in
two files, and then transferred to the evaluation plane. The information on Cursor key data for
the Bode plots can then be determined relatively accurately. In addition, there the amplitude
spectra of the two signals, in addition to this plot nor shown, as well as the SNR of these
signals. The generated files are saved in .txt format, the last two entries of this text file, the
start frequency and the end frequency include resolution to allow an evaluation with an Office
program.
3. Preparation Tasks
1. A sensor signal needs to be recorded. However, this is very weak and it has to
overcome a great distance from the sensor to the transmitter. List all known ways to
keep the noise as low as possible or to suppress in order to increase the SNR and thus
the detection accuracy.
2. Give typical filter characteristic and briefly describe essential properties.
3. Which filter characteristic does a 1st order filter have?
4. In a situation, a SNR of 25 dB is required. The useful signal has a frequency of 10 Hz
1 kHz. The transmitter generates a spectral noise voltage of 5 nV /Hz. The sensor
can be approximated for the noise analysis by a thermally active resistor 1k. How big
the useful signal must be to meet the given conditions at 20 C and 100 C? The noise
coupling can be neglected.
5. In order to suppress higher-frequency disturbances, a 2nd order low pass filter can be
used. Which filter characteristics (Bessel, Butterworth or Tschebyshev) would you
choose to influence as little as possible, the useful signal and to suppress interference?
Justify this decision! Calculate the cut-off frequency needed to obtain at 1 kHz at least
95% of the output amplitude, if the gain in the Pass band is A = 1 ! (H (j) | H (j)
| g)
6. Dimension the filter of 3.5. for MFB topology. C1 = 10 nF; C2 = 100 nF. Note sign!
7. Why no dedicated inductors are used in active filters?
8. How can you tell that a circuit and in particular a structured filter oscillates?
9. A weak, analog sensor signal to be digitized, processed and passed on as a control
signal to an actuator. The bandwidth of the signal is known. Which filters are
necessary and / or useful? Justify, at which point the signal path you would use (digital
/ analog) for any purpose that filter!
10. A low-pass first order is frequently used as an antialiasing filter for measurement
cards. Explain this observation. Dimension this filter for a sampling frequency of
125 kHz so that after sampling certainly no aliasing occurs! In order to preserve the
value of the required attenuation, the following data are needed:
Resolution: 16 bits
C=270pF
Is such a filter practicable? Explain!
4. Practical tasks
1. Familiarize yourself with the program and take a "no-load characteristic", the two
OPVs can be operated as a voltage follower! Use the white noise as a reference signal!
2. Determine a resistance value of the input filter, where the filter has the best influence
on the measurement! Explain!
3. Generate a sinusoidal signal with the measurement program! Check the signal with
and without input filter! (R = 100k, C = 10nF, C = 100 nF)
4. Inspect the influence of the circuit components. Vary the frequency between 100Hz
and 5 kHz with R = 100k and C = 10nF. Determine the cut-off frequency.
5. Build the filter from the Preparation Exercise 3.6. And assume its characteristic. Draw
the Bode plot without phase.
6. Make yourself familiar with the operation of the "Filter Free" program!
Let a Butterworth high pass 2nd order filter with a cutoff frequency fg = 2 kHz
get charged.
Build on the filter and compare the measurement result with the filter program!