Signal Recovery - Prof. S. Cova - Exam 2016/09/29 - P2 – pag.

PROBLEM 2

Data summary
RTD Thermoresistance PT100

Reference value at 0°C (273 K) RT0 =100 Ω

temperature coefficient α = 3,9∙10-3 /°C

Max dissipation Pd < 50µW

Si pn Junction Diode Sensors

  qV  
Current-Voltage Characteristics I  I S exp    1 with I s  0,1 pA =100 fA
  kT  

Max dissipation Pd< 50µW

Differential Preamplifier

Band-limit by a simple pole at fpa = 1 MHz

Effective noise power density

Sv1/2 = 80 nV/Hz1/2 white (unilateral) and 1/f component with fc = 20 kHz

Si1/2 = 1 pA/Hz1/2 white (unilateral) and 1/f component with fc = 20 kHz

(A) Sensor configuration for operation with DC voltage bias

A1) Thermoresistances

Wheatstone bridge with 4 equal resistances (1 RTD + 3 constant resistances) and DC voltage
supply VA .

Power dissipation in the RTD sensor

2
 VA  1
   Pd max = 50μW hence VA  2 Pd max RT 0  141mV
 2  RT 0
we select VA= 140 mV.

RT
Temperature variation ΔT produces resistance variation   T
RT 0
hence voltage variation at bridge output
VA RT VA
VS     T  136,5 V  T
4 RT 0 4
dVS
The conversion factor is  136,5 V K
dT
 Signal Recovery - Prof. S. Cova - Exam 2016/09/29 - P2 – pag.2

A2) Si pn junction diode Sensors

For a silicon pn junction biased at forward current I  I s  100 fA

  qV    qV  kT  I 
I  I S exp    1  I S exp   and V ln  
  kT    kT  q  IS 

A configuration with DC voltage bias employs two identical sensors submitted to the temperature T
to be measured and biased at two different DC current levels I1 e I2. The difference VD between the
forward voltages of the two diodes is read with a differential preamplifier

kT   I1   I 2   kT  I1 
VD  ln       ln  
q   IS   I S  q  I 2 
For evaluating the power dissipated in a diode, the forward bias voltage can be considered with
good approximation V= 600 mV.
Pd max
I V  I  600mV  Pd max = 50μW hence I  83  A
0,6 V

I1
Let us select I1 = 80 μA and I2 = 8 μA that is  10 .
I2

We get then

kT  I1  kT
VD  ln    2,3 that is, at room temperature T=300k we have VD ≈ 57,5 mV
q  I2  q

The conversion factor is

dVD k  I1  VD
 ln     192  V K
dT q  I2  T

(B) Noise and sensitivity limits in measurements with sensors operating with DC bias

The voltage noise SV of the preamplifier is the dominant noise contribution.

The contribution of the current noise Si of the preamplifier is negligible, because it is converted in
voltage noise by the low resistance RS of the signal source. In the case of thermoresistances, the RS
is the sensor resistance RT ≈ 100Ω . In the case of Silicon pn diodes the RS is the resistance of the
diode in forward bias
dV1 kT dV2 kT
RD1    312  RD 2    3,12 k 
dI1 qI1 dI 2 qI 2

therefore, in all cases RS Si1/2  10 nV Hz  Sv  80 nV Hz .

The intrinsic noise of the sensors (pn junctions or thermoresistances) can be directly verified to be
negligible.
 Signal Recovery - Prof. S. Cova - Exam 2016/09/29 - P2 – pag.3

In conclusion, the total noise to be taken into account is given by the voltage noise of the
preamplifier.
ST  f   Sv  f 

B1) Measurement without low-pass filtering and with CDS

The measurement is carried out with correlated double sampling CDS, setting to zero the amplifier
baseline in the available interval every Ti = 15-20 min ≈ 1000s .

The noise is subject to

low-pass filtering with limit f S   f pa 2  1,5MHz due to the preamp band-limit

The contributions of white noise and 1/f noise are

nB2  Sv1/2 2 f S  fi  2 Sv1/2 f S1/2  138V

f 
n 2f  2 Sv1/2 f C1/2 ln  S   74 V
 fi 
total noise

nT2  nB2  n 2f  156V  VS min

In this condition the white noise is dominant and the minimum measurable voltage variation ΔVS,min
corresponds to a temperature variation higher than the specification:
VS min
Tmin   1140mK with the RTD
dVS
dT
VD min
Tmin   810mK with the diode sensor
dVD
dT

B2) Measurement with low-pass filtering and CDS

A low-pass filter is employed at the preamplifier output for reducing the white noise contribution
and CDS as outlined in Sec.B1 is carried out on the filter output for reducing the 1/f noise
contribution. Temperature variations over time intervals down to 0,1 s must be measured, hence the
low-pass filter must pass frequency components up to a few 10 Hz; therefore, we employ a low-
pass with band-limit fS =100 Hz . We get

nB2  Sv1/2 2 f S  f i  2 Sv1/2 f S1/2  1,1V

f 
n 2f  2 Sv1/2 f C1/2 ln  S   54, 2 V
 fi 
 Signal Recovery - Prof. S. Cova - Exam 2016/09/29 - P2 – pag.4

the total noise is

nT2  nB2  n 2f  n 2f  54,2V  VS min

In this condition the 1/f noise is dominant and the minimum measurable voltage variation ΔVS,min
corresponds to a temperature variation which is reduced with respect to Sec. B1, but still higher
than the specification:
VS min
Tmin   400mK with the RTD
dVS
dT
VD min
Tmin   282mK with the diode sensor
dVD
dT

We note that the noise is so high because with DC bias the signal to be measured is in a low
frequency region, where the 1/f noise density is very high, much higher than the white noise
density.

(C) Measurement with sensor under time-varying bias (modulated or commuted voltage)

Employing a periodic time varying bias with fundamental frequency fm higher than the 1/f corner
frequency fc , the signal can be shifted to fm before entering in the preamplifier, i.e., before the
addition of 1/f noise. It is thus possible to employ a narrow-band filtering centered at fm for
collecting the signal accompanied by a reduced noise.

The two sensor have different features and will be dealt with separately.

C1) Thermoresistances RTD

The sensor is a linear resistor and we can simply employ a sinusoidal voltage supply with frequency
fm =100kHz and amplitude VA equal to the DC supply VA  140mV , thus maintaining the same

conversion factor (it would also be possible to employ an amplitude VA higher by a factor 2 , i.e.
VA  200mV without exceeding the dissipation limit).

A lock-in amplifier (LIA) with reference obtained from the voltage supply (which points out the
frequency and phase of the signal) can be employed for filtering the preamp output with a narrow
frequency band. The LIA demodulates the signal bringing it back to the base band (i.e. around f=0)
and filters it with its internal low-pass filter. For the reasons explained in Sec. B2, the band-limit of
this filter is set at fS =100 Hz.

The S/N obtained at LIA output is

as it can be shown in various ways (employing the LIA weighting function; analyizing the power
transfer of the various frequency components of the input and taking into account the selection in
frequency and phase by the LIA; etc.)

The minimum measurable amplitude variation (S/N=1) is

and the corresponding minmum measurable temperature variation is adequate to the specification

VS min
Tmin   8mK
dVS
dT

C2) Silicon pn junction Sensors

A sinusoidal current supply is not well suitable in this case. The pn junction has strongly nonlinear
characteristics, the modulated voltage waveform is strongly distorted with respect to the sinusoid of
the current and it is not simple to obtain the temperature value from the measured voltage signal
amplitude.

It is more suitable to employ bias current modulated at squarewave and obtain a squarewave voltage
signal with peak-to-peak amplitude VD  V1  V2 , with V1 and V2 corresponding to levels I1 and I2
in the diode, as shown in Sec.A2. It is then possible to measure directly the peak-to-peak amplitude
4 VD
VD  V1  V2 , or to measure the fundamental component of the squarewave, which is .
 2

Various current commutation schemes can be employed for producing the squarewave modulation.
In the scheme with two identical diodes seen in Sec.A2, the current can be commuted between I1
and I2 in a diode and kept constant at I2 in the other diode. In an alternative scheme that employs
just one diode, the current in the diode can be commuted between I1 and I2 .

We can employ a sinusoidal voltage signal generator at fm =100kHz and with a suitable switching
circuit obtain from it a square-wave signal for driving the current commutation in the diode.

We can then employ the sinusoidal signal as reference for a LIA that has the same settings as in
Sec.C1 and processes signal and noise coming from the preamp. In this case the LIA recovers only
the fundamental component at frequency fm , which has amplitude

4 VD 2VD

 2 
We get then
2VD
S
  V 2
 1/2D
N 2 Sv f S Sv  f S
 Signal Recovery - Prof. S. Cova - Exam 2016/09/29 - P2 – pag.6

and minimum measurable amplitude

VD min
Tmin   9mK
dVD
dT

If we instead employ the squarewave as reference for the LIA, then reference and signal have the
same waveform, the LIA exploits all the power of the signal and gives a better S/N
S VD V
  1/2 D
N 2 Sv f S Sv 2 f S

with improved minimum measurable amplitude

and correspondingly improved minimum measurable temperature variation

(D) Conclusions, summary and comment

It is required to measure small temperature variations (down to 10mK) occurring over fairly short
time intervals (down to 0,1 s). The analysis of the problem shows that

 For attaining the required goal, the contribution of the 1/f noise must be drastically limited.

 This result cannot be obtained by schemes employing stationary bias of the sensor, because
the signal produced is in the spectral region where 1/f noise is dominant.

 Satisfactory results are obtained by employing a time varying bias of the sensor (modulated
or commuted) which produces signals in a high-frequency spectral region where the 1/f
noise is negligible.

 In order to extract efficiently the signal from noise a narrow-band filtering must be
employed, with bandpass just wider than that of the signal, that is, such that the variations
in time of the temperature are faithfully transmitted

 The required narrow-band filtering can be obtained simply and efficiently with a LIA.