ELECTRONICS AND ELECTRICAL ENGINEERING

ISSN 1392 – 1215 2007. Nr. 4(76)

ELEKTRONIKA IR ELEKTROTECHNIKA

SIGNAL TECHNOLOGY
T 121
SIGNALŲ TECHNOLOGIJA

Digital Measurement of Frequency with Linear Interpolation

in Dynamic States
L. Referowski, D. Swisulski
Faculty of Electrical and Control Engineering, Gdansk University of Technology,
Narutowicza 11/12, 80-952 Gdańsk, Poland, phone: +48 58 3471726; e-mail: [email protected], [email protected]

Introduction preceding instant tk (Fig. 1).

Frequency modulated pulse signals are often used in

measuring channels [1] as they can be converted into
digital form in easy way and are more resistant against
disturbances especially in case of long distance
transmission [2]. A counter converts the analogue signal to
digital form therefore the resolution depends on the
capacity of the counter and the interval of measurement.
A counter with a great capacity, by selecting suitable
gating time, makes in easy way a transducer of 16 bits or
more. It is difficult to obtain transducers with the same
resolution in the case of voltage signal. The advantage of
f/C conversion in comparison with A/C conversion is the
possibility of easy access to standards of high accuracy. Fig. 1. Frequency measurement from two last periods
Digital frequency measurements of small values of
⎛ ⎛ i−2 ⎞⎞
frequency are based on the measurement of the period of ( fi − f i −1 )⎜ t k − ⎜⎜ ∑ T j + 12 Ti −1 ⎟⎟ ⎟
⎜ ⎟
the investigated signal. It can be determined by the ⎝ j =1 ⎠⎠
f k ' = f i −1 + ⎝ =
counting of pulses of the standard frequency generator in
⎛ i −1 ⎞ ⎛ i−2 ⎞
the time window defined by the measured period or its ⎜ ∑ T j + 12 Ti ⎟ − ⎜ ∑ T j + 12 Ti −1 ⎟
multiple [3]. The result of measurement corresponds to the ⎜ ⎟ ⎜ ⎟
⎝ j =1 ⎠ ⎝ j =1 ⎠
mean value of frequency in the defined time window. In
these measurements the instant which starts the ⎛1 1 ⎞⎛ i −2 ⎞
measurement depends on the position of the slope of the 2⎜⎜ − ⎟⎟⎜⎜ t k − ∑ T j − 12 Ti −1 ⎟⎟
1 ⎝ Ti Ti −1 ⎠⎝ j =1 ⎠.
measured signal. For this reason it is impossible to = + (1)
compare the results of measurements in different Ti −1 Ti −1 + Ti
measuring channels especially frequency and voltage
channels. When the frequency of the signal changes in The formula (1) is valid in assumption that the
time of measurement the results are delivered in different frequency varies in linear mode i.e. the computed value of
intervals and therefore it is impossible to calculate Fourier frequency is equal to the value of frequency in the centre
transform. In this case it is recommended to use the linear of examined period.
interpolation. In the paper are presented two such methods. Practically two counters are used for the
measurement. The first one works in a buffer mode and
On-line mode measures intervals between the pulses of the same signal.
Therefore for arbitrary instant in the memory are registered
The first method is used in an on-line mode [4]. When all previous intervals between pulses of measured signal.
the value of frequency is determined in instant tk one has The second counter registered the intervals of pulses of
only information of instants pulses preceding instant tk. different signals. It measures interval Tk between the last
The value of frequency fk in the instant tk, can be pulse of pulse signal and the pulse determining the instant
calculated as f’k (formula 1) by extrapolation the periods of measurement tk. The value of frequency in the instant of

47
tk is calculated basing on the values of last periods Ti-1 and ⎛ i ⎞
Ti and interval Tk according to the formula 2( f i + 2 − f i +1 )⎜⎜ t k − ∑ Tj − 12 Ti +1 ⎟⎟
⎝ j =1 ⎠. (4)
= f i +1 +
⎛1 1 ⎞1 Ti +1 + Ti + 2
2⎜⎜ − ⎟( 2 Ti −1 + Ti + Tk )
1 T Ti −1 ⎟⎠
fk '= + ⎝ i =
Ti −1 Ti −1 + Ti For one measuring channel with frequency carrier
complete measuring cycle is realised by two counters
T ⋅ (T + 2Ti + 2Tk ) − Ti ⋅ (Ti + Tk ) . (2)
= i −1 i −1 working in continuous buffered mode. The first one counts
Ti −1 ⋅ Tk ⋅ (Ti −1 + Ti ) pulses of standard signal in meantime of consecutive
periods of pulse signal, and loads them to the memory. The
Off-line mode second counter counts the pulses of standard signal in the
instants between the pulse determining the instant of
The second method is used in an off-line mode [5]. In sampling and the next pulse of measured signal.
the first step one defines and registers the positions of In the instant of measurement the number of last
pulses of measured signal. Then the value of frequency is measured period is registered in the memory (the number
computed for chosen instants. For the arbitrary chosen of measured periods up to the instant of given sampling
instant of frequency the position of pulses preceding and pulse).
following this instant is well known. After the end of measuring cycle in the memory are
The value of frequency fk in the instant tk can be registered three one dimensional table for simple
calculated from two successive periods in assumption that measuring channel:
the frequency has changed in linear mode. Depending on • numbers of pulse signal in instants of strobe pulses: a1,
the instant of tk these periods are Ti and Ti+1 for calculation a2, …, am,
frequency fk or Ti+1 and Ti+2 for frequency f’k (Fig. 2). • length of periods of the measured signal: T1, T2, …, Tn,
• intervals between the sampling pulse and the nearest
pulse of measured signal: τ1, τ2, …, τp.
The first two tables have the same numbers of
elements m. The third table has the same number of
elements as the two first tables (p = m) in the case when
the longest period of pulse signal is equal or less than the
periods of sampling (max Ti ≤ Ts) as presented in Fig. 3.

Fig. 2. Frequency measurement from two periods

Fig. 3. Sampling for max Ti ≤ Ts
i i
In the case ∑T
j =1
j ≤ tk ≤ ∑ T j + Ti +1
j =1
1
2
frequency fk
Whereas, if the longest period of pulse signal is
(Fig. 2) is calculate according to the formula greater than the sampling period (max Ti > Ts), the third
table could have less elements than the two first (p ≤ m). It
⎛ ⎛ i −1 ⎞⎞ is the result of working mode of counters measuring
( f i +1 − f i )⎜⎜ t k − ⎜⎜ ∑ Tj + 12 Ti ⎟⎟ ⎟⎟ intervals between pulses of different signals. After the
⎝ ⎝ j =1 ⎠⎠ pulse of the first signal initiating the measurement the
fk = fi + =
⎛ i ⎞ ⎛ i −1 ⎞ measuring cycle is finished in the instant of the first pulse
⎜ ∑ Tj + 12 Ti +1 ⎟ − ⎜ ∑ Tj + 12 Ti ⎟
⎜ ⎟ ⎜ ⎟ of the second signal regardless of the successive pulses that
⎝ j =1 ⎠ ⎝ j =1 ⎠ can appear in the first signal (Fig. 4).
⎛ i −1 ⎞
2( f i+1 − f i )⎜⎜ t k − ∑ Tj − 12 Ti ⎟⎟
= fi + ⎝ j =1 ⎠, (3)
Ti + Ti+1
i i +1
while in the case ∑ T j + 12 Ti +1 < t k < ∑ T j frequency f’k
j =1 j =1 Fig. 4. Sampling for max Ti > Ts
(Fig. 2) is calculated according to the formula
Omitted intervals τj between both the signals must be
⎛ ⎛ i ⎞⎞ taken into consideration, therefore in the third table in the
( f i + 2 − f i +1 )⎜⎜ t k − ⎜⎜ ∑ Tj + 12 Ti +1 ⎟⎟ ⎟⎟ place of individual element τj for τj > Tpr w elements τ’1,
⎝ ⎝ j =1 ⎠⎠
f ' k = f i +1 + = τ’2, …, τ’w are inserted (Fig. 5), where:
⎛ i + 1 ⎞ ⎛ i ⎞ τ '1 = τ j , τ ' 2 = τ j − Ts , τ ' w = τ j − (w − 1)Tw , (5)
⎜ ∑ Tj + 2 Ti + 2 ⎟ − ⎜ ∑ Tj + 2 Ti +1 ⎟
1 1
⎜ ⎟ ⎜ ⎟
⎝ j =1 ⎠ ⎝ j =1 ⎠

48
 ⎡τ j ⎤ where f0 – constant component of frequency of pulse signal
w = trunc ⎢ ⎥ + 1 , (6) f, fm – amplitude of deviation of pulse signal frequency f.
⎣ Ts ⎦ i −1
For the instant t i = ∑ T j + 1Ti the value of frequency
2
j =1
and trunc[x] is integer of number x.
is expressed as

⎡ ⎛ i −1 ⎞⎤
f i = f 0 + f m ⋅ sin ⎢2πf sin ⎜⎜ ∑ T j + 12Ti ⎟⎟⎥ + noisei . (10)
⎣⎢ ⎝ j =1 ⎠⎦⎥
After transformation equation (11) is obtained in form

⎡ ⎛ i −1 T ⎞⎤ 1
f 0 + f m ⋅ sin ⎢2πf sin ⎜⎜ ∑ T j + i ⎟⎟⎥ + noisei − = 0 , (11)
⎣⎢ ⎝ j =1 2 ⎠⎦⎥ Ti
Fig. 5. Conversion of table τp
From this equation it is possible to compute the
In that way table τ1, τ2, … , is created τm with m values of successive intervals between the pulses Ti of
elements, where successive values represent interval measured signal.
between every sampling pulse and the nearest pulse of In the Fig. 6 distribution of errors is presented for
pulse signal. sinusoidal fluctuation of the pulse signal in the case of
The value of frequency fk of pulse signal in the f0 = 20 Hz, fm = 2 Hz, fsin = 0,3 Hz, noisemax = 0,001 Hz
instant of kth pulse of sampling signal is computed from and sampling frequency fs = 50 Hz.
formulae (3) and (4) taking into consideration two periods
( Ta i Ta +1 or Ta +1 i Ta + 2 ) and interval τ a , where ak is
k k k k k

index of period which precedes sampling signal.

For τ a ≥ 12 Ta +1 frequency fk is calculated from
k k

formula

⎛ 1 1 ⎞⎟ 1
⎜
⎜
− (
Ta + Tak +1 + τ ak
Tak +1 Tak ⎟⎠ 2 k
)
1 ⎝
fk = + =
Tak 1
T + 12 Tak +1
2 ak

=
(
Tak 2 − Tak +1 2 + 2 ⋅ Tak ⋅ Tak +1 − 2 ⋅ τ ak ⋅ Tak + Tak +1 ) . (7) Fig. 6. Errors Δfk distribution for f0 = 20 Hz, fm = 2 Hz,
fsin = 0,3 Hz, noisemax = 0,001 Hz and fs = 50 Hz;
(T
ak )
+ Tak +1 ⋅ Tak ⋅ Tak +1 a – measurement from last time interval, b – measurement from
two last intervals
While for τ a < 12 Ta +1 frequency fk is used formula
k k

⎛ 1 1 ⎞⎟ 1
⎜ − T
⎜ Ta + 2 Ta +1 ⎟ 2 ak +1
( − τ ak )
1 ⎝ ⎠
f 'k = + k k
=
Tak +1 2 Ta k +1 + 2 Ta k + 2
1 1

=
(
Tak +1 2 − Tak + 2 2 − 2 ⋅ τ ak ⋅ Tak +1 − Tak + 2 ). (8)
(T ak +1 )
+ Tak + 2 ⋅ Tak +1 ⋅ Tak + 2

In this way one dimensional table is created for the

specific channel with the values of frequency in the Fig. 7. Errors Δfk distribution for f0 = 20 Hz, fm = 2 Hz,
instants of sampling: f1, f2, …, fm. fsin = 0,3 Hz, noisemax = 0,1 Hz and fs = 50 Hz; a – measurement
from last time interval, b – measurement from two last intervals
The numbers of investigated channels can be greater
than one and is only restricted by the numbers of counters
The distribution of errors in the case of the same
(there must be two counters for individual channel).
parameters as in Fig. 6 but for noisemax = 0,1 Hz is
presented in the Fig. 7.
Simulation of frequency measurements
Comparing Figs. 6 and 7 one can ascertain that for
the linear changes of frequency the method of computing
Consider a measured signal f expressed by formula
the frequency from two last pulses intervals enables to
(9) that fluctuates in sinusoidal way with a frequency fsin,
obtain good results in case of small changes of measured
immersed in a noise in the time of measurement [4]
frequency. For the great values of noise this method can
f = f 0 + f m ⋅ sin (2πf sin t ) + noise , (9) even worsen the accuracy of measurements.

49
Conclusions 2. Świsulski D. Cyfrowa rejestracja sygnałów impulsowych z
częstotliwościowym nośnikiem informacji. – Gdańsk
In the paper the principles and realisation of pulse Wydawnictwo Politechniki Gdańskiej, 2006.
signal frequency measurement are presented in dynamic 3. Prokin M. Dynamic Response of a Frequency Measuring
System. IEEE Transactions on Instrumentation and
condition. The presented methods use linear interpolation. Measurements. – Vol. 41, No. 3, June 1992. – P. 390–396.
From the simulation of measurements for sinusoidal 4. Świsulski D. Pomiar on-line zmiennej częstotliwości metodą
variations of measured signal the interpolation method cyfrową // Pomiary Automatyka Kontrola. – 2006, No. 9bis. –
gives good results for small amount of noises in the P. 101–103.
measured signal. 5. Świsulski D. Wielokanałowa akwizycja z torami
pomiarowymi z napięciowym i częstotliwościowym
References nośnikiem informacji. Pomiary Automatyka Kontrola. –
2006. – No. 6. – P. 27-29.
1. Kirianaki N. V., Yurish S. Y., Shpak N. O., Denega V. P.
Data acquisition and signal processing for smart sensors. –
John Wiley & Sons, Ltd, Baffins Lane 2001. Submitted for publication 2007 02 15

L. Referowski, D. Swisulski. Digital Measurement of Frequency with Linear Interpolation in Dynamic States // Electronics and
Electrical Engineering – Kaunas: Technologija, 2007. – No. 4(76). – P. 47–50.
Digital frequency measurements of small values of frequency are based on the measurement of the period of investigated signal. The
result of measurement corresponds to the mean value of frequency in the defined time window. The instants which start the
measurement depend on the position of the slope of the measured signal. For this reason it is impossible to compare the results of
measurements in different measuring channels or to compare the results from frequency channels and voltage channels. The results are
delivered in different intervals and therefore it is impossible to calculate Fourier transform. Good results can assure the linear
interpolation. In the paper two methods (on-line and off-line mode) realized with buffered counters are presented. The simulation proves
that good results of measurements can be obtained in case of a little amount of noises in the measured signal. Ill. 7, bibl. 5 (in English;
summaries in English, Russian and Lithuanian).

Л. Реферовский, Д. Свисульский. Цифровые измерения частоты с линейной интерполяцией в динамических истенах //

Электроника и электротехника. – Каунас: Технология, 2007. – № 4(76). – С. 47–50.
Цифровые измерения сигнала для малых величин частоты, основаны на измерении его периода. Результат измерения – это
средняя величина частоты в определённом пределе часа. Моменты, в которых начинается измерение, зависят от позиции
склона измеренного сигнала. Поэтому невозможо сравнить результаты измерения в разных каналах или результаты измерений
в каналах частоты и напряжения. Результаты получаются в разных периодах и поэтому невозможно вычислить
трансформацию Фурье. Хорошие результаты возможно получить используя линейную интерполяцию. В статье представлены
два метода (он-лан и офф-лан) связанные с буферными счётчиками. Результаты симуляции показывают, что хорошие
результаты измерения можно получить при малом шуме в исследованном сигнале. Ил. 7, библ. 5 (на английском, русском и
литовском яз.).

L. Referowski, D. Swisulski. Skaitmeninis dažnio matavimas naudojant tiesinę interpoliaciją dinaminėse sistemose //
Elektronika ir elektrotechnika. – Kaunas: Technologija, 2007. – Nr. 4(76). – P. 47–50.
Išnagrinėtas skaitmeninio signalo matavimas, esant mažoms dažnio reikšmėms, pagrįstas tiriamojo signalo periodo matavimu.
Matavimo rezultatą atitinka vidutinė dažnio reikšmė tam tikru laiko momentu. Momentai, kuriais prasideda matavimai, priklauso nuo
matuojamojo signalo formos. Dėl šios priežasties neįmanoma palyginti skirtinguose kanaluose arba dažnio ir įtampos kanaluose atliktų
matavimų rezultatų. Rezultatai gaunami ir apdorojami skirtingais laiko momentais, tačiau Furjė transformacijos neįmanoma
apskaičiuoti. Naudojant tiesinę interpoliaciją galima tikėtis gerų rezultatų. Išnagrinėti du metodai, susiję su buferiniais skaitikliais.
Modeliavimo rezultatai rodo, kad geriausi matavimo rezultatai gaunami tik tada, kai matuojamame signale yra mažai triukšmų. Il. 7,
bibl. 5, (lietuvių kalba; santraukos anglų, rusų ir lietuvių k.).

50