The Output Resistance of a power supply

Keegan Humphrey and Muhammad Aazmeer Asif

October 28th 2019

1 Introduction
Approximating zero resistance for the power supply, multimeters, or wiring of a circuit is commonplace among
undergraduates. This allows us to simplify our calculations greatly and serves as a powerful tool, with the
caveat of some uncertainty introducued in our readings. In this lab we explored the internal resistance of a
power supply and some resistors. Throughout our report we hope to diagnose this uncertainty by comparing
the open circuit voltage (V∞ ), with the terminal voltage (V ) - when all the components are attached. This
terminal voltage is expressed as:
V = V∞ − RI (1)
Where R is the internal resistance of our power supply or battery, and I is the current running through the
ammeter. Hence, we expect the terminal voltage to always be less than its open circuit counterpart. In the
following sections we explore this clause using 2 different circuit setups with both a battery (fixed voltage),
and a power supply (variable voltage).

Figure 1: Our 2 Circuit setups

2 Materials and Methods

In this lab we used the following materials:
• Power supply
• Battery (6.5 V)
• 2 Tegam Multimeters
• Resistors (680, 820, 8200, 33000 Ω) - each resistor having an uncertainty of ± 5% in its value
• Connecting wires
• Computer with Python v3.7 installed
We used the power supply and battery for 2 different circuit options as shown in Figure 1. We first attached
the battery to the options and recorded the open circuit voltage in each which came out to be 6.42 V. Table
1 below shows our measured V and I values for the battery.
We then conducted the same process with the power supply at multiple open circuit voltages (6.42, 10.02,
14.99, 20.00 V). This data is subsequently presented in the tables on the next page.

1
Resistance (Ω) Option 1 - Voltage (V) Current (mA) Option 2 - Voltage (V) Current (mA)
680 6.41 8.99 6.31 8.99
820 6.41 7.67 6.33 7.67
8200 6.41 0.761 6.34 0.762
33000 6.41 0.195 6.39 0.196

Table 1: V and I measurements for the battery

Resistance (Ω) Option 1 - Voltage (V) Current (mA) Option 2 - Voltage (V) Current (mA)
680 6.37 8.93 6.28 8.94
820 6.37 7.62 6.30 7.63
8200 6.38 0.757 6.38 0.764
33000 6.38 0.194 6.39 0.192

Table 2: Readings for the power source - V∞ =6.42 V

Resistance (Ω) Option 1 - Voltage (V) Current (mA) Option 2 - Voltage (V) Current (mA)
680 10.05 14.11 9.83 14.00
820 10.06 12.05 9.86 11.94
8200 10.08 1.196 9.88 1.187
33000 10.08 0.307 10.00 0.305

Table 3: Readings for the power source - V∞ =10.09 V

Resistance (Ω) Option 1 - Voltage (V) Current (mA) Option 2 - Voltage (V) Current (mA)
680 14.93 21.3 14.70 21.0
820 14.94 17.92 14.75 17.93
8200 14.97 1.776 14.79 1.777
33000 14.98 0.456 14.93 0.458

Table 4: Readings for the power source - V∞ =14.99 V

Resistance (Ω) Option 1 - Voltage (V) Current (mA) Option 2 - Voltage (V) Current (mA)
680 19.9 28.5 19.8 28.6
820 19.9 24.4 19.9 24.4
8200 19.9 2.40 19.7 2.40
33000 19.9 0.611 19.9 0.614

Table 5: Readings for the power source - V∞ =20.0 V

2
3 Data Analyses
We then transcribed (1) into Python, using it to plot our data. The function we used to model (1) is as
follows:
f (x, a, b) = a ∗ x + b (2)
Where a is -R and b is open circuit voltage (Vinf ). The uncertainty in our data, though not shown in our
tables - due to the extreme hassle of making said tables, was selected between precision and accuracy types.
The precision uncertainty in each reading was simply taken to as ± the smallest significant figure in it.
Furthermore, the accuracy error was derived from the Tegam user manual as a percentage of the absolute
reading. This varied based on which settings we were using. We will further discuss the consequence
of actively changing the uncertainty via the settings in the discussion; refer to the appendix for explicit
uncertainty calculations. We conclude this section by presenting our plotted results in the next several
figures.

Figure 2: Current vs Potential for the battery - Vinf =6.42V

In order to analyze the goodness of our fit we transcribed the χ2 test into Python, which states:
N
1 X yi − y(xi ) 2
χ2 = ( ) (3)
v i=1 σ

Where v is the degrees of freedom, N is the number of samples, y is our data, σ is our uncertainty in y, and
y(x) is our model function. From the handout we also know that a χ2 value less than 1 indicates a poor fit,
while a value greater than 1 indicates an overfit.

4 Discussion
We first start by comparing the 2 different options we had for the circuit (Figure 1). In option 1 we see
that the ammeter is attached in a way such that the voltmeter measures the voltage drop across both it and

3
Figure 3: Current vs Potential for option 1 set with the power supply, each data point represents readings
taken for each chosen resistor

our external resistor (Ri ). In option 2 the voltmeter only measures the voltage drop across the resistor. If
we assume ideal multimeters (0 resistance for the ammeter, ∞ resistance for the voltmeter), this change in
position does not matter. However our instruments are less than ideal and hence some current is bound to
flow through the voltmeter and some voltage is bound to be dropped by the ammeter. From our figures we
can see that option 1 is better suited for our experiment. To understand why we can reiterate the goal of
the experiment which was to calculate the internal resistance of the power supply/battery, hence bundling
all of the external components of our circuit together and measuring the voltage drop across them is a better
representation of ’IR’ in equation (1). Secondly we take note of our choice of external resistors. This was
done simply to maximize our range of values from all of the possible resistors given to us. This selection
proved to be an unwise choice as we will see later.
Using the tables above we can derive the current flowing through the voltmeter for each value of V∞
and each resistor. An example of this calculation can be done using Table 2’s first row (680 Ω), the current
through the voltmeter in this case is the difference between the current measured for both options:

IV oltmeter = IOption2 − IOption1

IV oltmeter (680Ω, V∞ = 6.42V ) = 8.94 − 8.93 = 0.01mA

Similarly we can compare both options to get a gauge of the ammeter’s resistance as well. As we can see the
terminal voltage changes between options, confirming that our ammeter is not ideal. We can then calculate
the resistance of the ammeter as:
VOption1 − VOption2
RAmmeter =
I
The current I needs to remain constant here so we can use table values which preserve this.

4
Figure 4: Current vs Potential for option 2 set with the power supply, each data point represents readings
taken for each chosen resistor

We can now proceed to analyze our graphs and calculate the resistances for both the batter and power
supply. Since we used curve fit to model our current vs potential, we can directly calculate the resistance
by using the p opt matrix generated by curve fit. Our parameter a is simply -R. The uncertainty in our
resistance was also subsequently calculated using the uncertainties in V and I. These calculated resistances
are shown in the tables below:

Power Option 1 - Resistance (Ω) Option 2 - Resistance (Ω)

Battery 8.480 6.052±0.012
Power Source (33k Ω) 1.261±0.002 1.226±0.002
Power source (8.2k Ω) 2.061±0.004 8.139±0.016
Power Source (820 Ω) 2.189±0.004 7.700±0.015
Power source (680 Ω) -2.749*10−7 ±-3.722*10−11 -1.355± -0.005

Table 6: Calculated resistances for our setups

The first, and perhaps more apparent thing, is that there is an enormous discrepancy in the chi squared
values for option 1 and option 2. In option 1 our chi squared values are fairly reasonable; they are within
an order of magnitude of 1 for a few of the voltages with one very small value of around .02. This is to be
expected for only 4 data points in each run of the regression. As a value of less than 1 for the chi squared
test indicates there are not enough data points or the model is fitting noise. Perhaps the next most obvious
thing is that we quoted a chi squared value of 0 for option 1 using the battery. This is because the values of
potential we measured were identical for each of the resistors. This led our regression to return a covariance
matrix with all entries equal to infinity. And since the chi square is inversely proportional to the variance, it
outputted the value of zero, and gave us an infinite error in the output resistance. For option 2, we returned
chi squared values multiple orders of magnitude larger than 1 (save one value). This is extremely surprising
considering the opposite was true of the option 1 data. We would expect that the values be close, or at

5
 Power Option 1 - χ2 Option 2 - χ2
Battery - 638124321
Power Source (33k Ω) 0.0285 8.08242
Power source (8.2k Ω) 0.3206 1378.3
Power Source (820 Ω) 0.5148 959.98
Power source (680 Ω) 1.9914 279.73

Table 7: Calculated χ2 values for our setups

least be on the same side of the inequality with 1. Upon review of the raw data we notice that the variation
between the two data sets is slight, making it seem unlikely that the result of the chi squared is due to
the variation between the data sets. The next most plausible explanation is an error in the script used to
perform the regression and chi squared test. However, both data sets were run together, as close as makes
no difference to identically. After extensive review it remains the most likely culprit, but the specific error
is not easily identified or explained.
It is also possible that due to the lack of a sufficient number of samples the regression and chi squared
became extremely sensitive to the random error and so produced large variations in the output values that
depend almost exclusively on the noise and random error. This final option seems very plausible when we
compare the errors across the two configurations. The error for option 2 is larger by about a factor of 10 for
most of the data. Which is exactly the opposite one would expect. As having a chi squared value greater
than 1 is generally associated with an underestimated error variance. Which is likely true for both options;
the error in the option 1 data is less than 0.5% for most of the values quoted for the resistance.
We also returned an exceptionally small value (10 orders of magnitude smaller than 1) for the chi squared,
and that same data returned an exceptionally small value for the error in the resistance. There again seems
to be no justifiable explanation except an unidentified error in the script used to calculate the value. What
can be said with near certainty is that the linear regression function produced a very poor fit for a majority
of the data. Perhaps it is due to the data being nonlinear (a departure predicted at large values of current),
or simply a result of an insufficient number of data samples. Another crucial mistake in our experiment
extends from our choise of external resistors. Due to such a large range our Current and Potential values
laid within range of multiple multimeter settings. In the interest of maximizing precision we changed the
settings multiple times within each set of readings. However we reamined ignorant to the fact that doing
so changes the accuracy of our readings. As a result we can also associate discontinuities within our results
due to that added error as well.

5 Conclusion
In this experiment we sought to explore the internal resistance of 2 power sources: a battery and a variable
power source. We found that while it may serve to be a good approximation, our power sources and
multimeters are not ideal. We expected the voltage to drop with increasing current, explained by the fact
that a lower overall resistance increases the current but lowers the terminal voltage (a greater share of the
output voltage goes to the power source due to its internal resistance). Throughout our experiment we
were not able to quantitatively establish this hypothesis, likely due to the errors and uncertainties in our
experiment. However, we were able to qualitatively explain the trend through our plots.

6
Appendix for the rawcode used
import numpy a s np
import s c i p y . o p t i m i z e a s s c i
import m a t p l o t l i b . p y p l o t a s p l t

R1 , V1 , I 1 = np . l o a d t x t ( ’ . / r e s i s t o r data ( 1 ) . txt ’ , unpack=True , s k i p r o w s =2)

R2 , V2 , I 2 = np . l o a d t x t ( ’ . / r e s i s t o r data ( 2 ) . txt ’ , unpack=True , s k i p r o w s =2)
r1 , i 1 , v1 = np . l o a d t x t ( ’ . / Option 1 . txt ’ , unpack=True , s k i p r o w s =3)
r2 , i 2 , v2 = np . l o a d t x t ( ’ . / Option 2 . txt ’ , unpack=True , s k i p r o w s =3)

V inf = [6.42 ,6.42 ,6.42 ,6.42 ,10.02 ,10.02 ,10.02 ,10.02 ,14.99 ,14.99 ,14.99 ,14.99...
...20.0 ,20.0 ,20.0 ,20.0]
Vinf = [ V inf [ 0 ] , V inf [ 4 ] , V inf [ 8 ] , V inf [ 1 2 ] ]

def f (x , a , b ) :
r e t u r n a∗x+b

d e f c h i (N, n , exp x , exp y , func , f u n c a r g 1 , f u n c a r g 2 , e r r ) :

r e t u r n np . sum ( ( ( exp y−f u n c ( exp x , f u n c a r g 1 , f u n c a r g 2 ) ) / e r r ) ∗ ∗ 2 ) / (N−n )

err1 = np . z e r o s ( 1 6 )
err2 = np . z e r o s ( 1 6 )
err3 = np . z e r o s ( 4 )
err4 = np . z e r o s ( 4 )

f o r i in range ( 1 6 ) :
e r r 1 [ i ] = np . max ( [ . 0 1 , . 0 0 2 5 ∗ V1 [ i ] ] )
e r r 2 [ i ] = np . max ( [ . 0 1 , . 0 0 2 5 ∗ V2 [ i ] ] )

f o r i in range ( 4 ) :
e r r 3 [ i ] = np . max ( [ . 0 0 1 , . 0 0 2 5 ∗ v1 [ i ] ] )
e r r 4 [ i ] = np . max ( [ . 0 0 1 , . 0 0 2 5 ∗ v2 [ i ] ] )

err1 [ −4:] = [ . 1 , . 1 , . 1 , . 1 ]
err2 [ −4:] = [ . 1 , . 1 , . 1 , . 1 ]

f4par , f4cov = sci . curve fit (f , i 1 , v1 , sigma=e r r 3 , a b s o l u t e s i g m a=True )

f3par , f3cov = sci . curve fit (f , I 1 [ 0 : 4 ] , V1 [ 0 : 4 ] , sigma=e r r 1 [ 0 : 4 ] , a b s o l u t e s i g m a=True )
f2par , f2cov = sci . curve fit (f , I 1 [ 4 : 8 ] , V1 [ 4 : 8 ] , sigma=e r r 1 [ 4 : 8 ] , a b s o l u t e s i g m a=True )
f1par , f1cov = sci . curve fit (f , I 1 [ 8 : 1 2 ] , V1 [ 8 : 1 2 ] , sigma=e r r 1 [ 8 : 1 2 ] , a b s o l u t e s i g m a=True )
f0par , f0cov = sci . curve fit (f , I 1 [ 1 2 : 1 6 ] , V1 [ 1 2 : 1 6 ] , sigma=e r r 1 [ 1 2 : 1 6 ] , a b s o l u t e s i g m a=True )

g4par , g4cov = sci . curve fit (f , i 2 , v2 , sigma=e r r 4 , a b s o l u t e s i g m a=True )

g3par , g3cov = sci . curve fit (f , I 2 [ 0 : 4 ] , V2 [ 0 : 4 ] , sigma=e r r 2 [ 0 : 4 ] , a b s o l u t e s i g m a=True )
g2par , g2cov = sci . curve fit (f , I 2 [ 4 : 8 ] , V2 [ 4 : 8 ] , sigma=e r r 2 [ 4 : 8 ] , a b s o l u t e s i g m a=True )
g1par , g1cov = sci . curve fit (f , I 2 [ 8 : 1 2 ] , V2 [ 8 : 1 2 ] , sigma=e r r 2 [ 8 : 1 2 ] , a b s o l u t e s i g m a=True )
g0par , g0cov = sci . curve fit (f , I 2 [ 1 2 : 1 6 ] , V2 [ 1 2 : 1 6 ] , sigma=e r r 2 [ 1 2 : 1 6 ] , a b s o l u t e s i g m a=True )

rl4 = −f 4 p a r [0]
rl3 = −f 3 p a r [0]
rl2 = −f 2 p a r [0]
rl1 = −f 1 p a r [0]
rl0 = −f 0 p a r [0]

7
Rl4 = −g4par [ 0 ]
Rl3 = −g3par [ 0 ]
Rl2 = −g2par [ 0 ]
Rl1 = −g1par [ 0 ]
Rl0 = −g0par [ 0 ]

delrl4 = np . sqrt ( f4cov [ 0 ,0])

delrl3 = np . sqrt ( f3cov [ 0 ,0])
delrl2 = np . sqrt ( f2cov [ 0 ,0])
delrl1 = np . sqrt ( f1cov [ 0 ,0])
delrl0 = np . sqrt ( f0cov [ 0 ,0])

delRl4 = np . s q r t ( g4cov [ 0 ,0])

delRl3 = np . s q r t ( g3cov [ 0 ,0])
delRl2 = np . s q r t ( g2cov [ 0 ,0])
delRl1 = np . s q r t ( g1cov [ 0 ,0])
delRl0 = np . s q r t ( g0cov [ 0 ,0])

plt . f i g u r e ( f i g s i z e =(10 ,8))

plt . subplot (2 ,1 ,1)
plt . plot ()
plt . e r r o r b a r ( i 1 , v1 , y e r r=e r r 3 , fmt = ’ ’ , l a b e l =’ b a t t e r y ’ )
plt . e r r o r b a r ( I 1 [ 0 : 4 ] , V1 [ 0 : 4 ] , y e r r=e r r 1 [ 0 : 4 ] , fmt = ’ ’ , l a b e l =’ V i n f = 6 . 4 2 ’ )
plt . e r r o r b a r ( I 1 [ 4 : 8 ] , V1 [ 4 : 8 ] , y e r r=e r r 1 [ 4 : 8 ] , fmt = ’ ’ , l a b e l =’ V i n f = 1 0 . 0 2 V’ )
plt . e r r o r b a r ( I 1 [ 8 : 1 2 ] , V1 [ 8 : 1 2 ] , y e r r=e r r 1 [ 8 : 1 2 ] , fmt = ’ ’ , l a b e l =’ V i n f = 1 4 . 9 9 V’ )
plt . e r r o r b a r ( I 1 [ 1 2 : 1 6 ] , V1 [ 1 2 : 1 6 ] , y e r r=e r r 1 [ 1 2 : 1 6 ] , fmt = ’ ’ , l a b e l =’ V i n f = 2 0 . 0 V’ )
plt . t i t l e ( ’ Current vs . P o t e n t i a l − Option 1 C i r c u i t ’ )
plt . x l a b e l ( ’ Current (mA) ’ )
plt . y l a b e l ( ’ P o t e n t i a l (V) ’ )
plt . legend ()
plt . ylim ( 0 , 2 5 )

plt . subplot (2 ,1 ,2)

plt . plot ()
plt . e r r o r b a r ( i 2 , v2 , y e r r = e r r 4 , fmt = ’ ’ , l a b e l =’ b a t t e r y ’ )
plt . e r r o r b a r ( I 2 [ 0 : 4 ] , V2 [ 0 : 4 ] , y e r r = e r r 2 [ 0 : 4 ] , fmt = ’ ’ , l a b e l =’ V i n f = 6 . 4 2 V’ )
plt . e r r o r b a r ( I 2 [ 4 : 8 ] , V2 [ 4 : 8 ] , y e r r = e r r 2 [ 4 : 8 ] , fmt = ’ ’ , l a b e l =’ V i n f = 1 0 . 0 2 V’ )
plt . e r r o r b a r ( I 2 [ 8 : 1 2 ] , V2 [ 8 : 1 2 ] , y e r r = e r r 2 [ 8 : 1 2 ] , fmt = ’ ’ , l a b e l =’ V i n f = 1 4 . 9 9 V’ )
plt . e r r o r b a r ( I 2 [ 1 2 : 1 6 ] , V2 [ 1 2 : 1 6 ] , y e r r = e r r 2 [ 1 2 : 1 6 ] , fmt = ’ ’ , l a b e l =’ V i n f = 2 0 . 0 V’ )
plt . t i t l e ( ’ Current vs . P o t e n t i a l − Option 1 C i r c u i t ’ )
plt . x l a b e l ( ’ Current (mA) ’ )
plt . y l a b e l ( ’ P o t e n t i a l (V) ’ )
plt . legend ()
plt . ylim ( 0 , 2 5 )