Rather then debate this subject aimlessly on the web, I decided to
spend some time getting some actual results. My previous experience
going back too many years had me convinced that as frequency was
pushed up the transformer losses increased. There must have been
some basis for that, so I better have a look in some detail in order
to satisfy my curiosity. Was it core losses or was it something
else?

All of these tests were made on a recent production Hammond 1650N.
It is spec'd for 60 watts audio in PP & has a 4300 ohm primary with
40 % UL taps. The secondary is the usual Hammond practice of three
windings brought out to provide match to 4, 8 & 16 ohm loads. The 8
ohm tap uses all of these windings while the 4 & 16 ohm connexions
use only 87 % of the secondary copper. The impedance ratio, primary
to secondary is 537.5. I used that figure for the calculation of the
primary parasitic resistance (Rpri in the table below).

I had previously posted similar sets of data measured from a 50 year
old Hammond 1700 Series OPT & a recent production Hammond 125E
Universal OPT. Some of the posts from others got me suspicious of my
test setup. I was using a Metra-Hit 29S Precision DMM which has the
ability to measure power, both real & reactive & Power Factor (PF)
along with all the other data. The RMS responding VM in the 29S DMM
is spec'd to 100 KHz. However, for power the published spec goes
only to one KHz. I had tried using it to 10 KHz & got some results
that were at least believable. As anyone who has ever tried knows,
the problem of making an AC current shunt of low resistance & large
bandwidth is daunting to say the least. Skin Effect is the culprit.
That is why you will not normally find AC current measurement to any
great frequency in a lower cost DMM. It is also the reason why the
power measurement in the 29S is spec'd only to one KHz. Another
consideration is 'how should the connexion be made?' If one is not
careful you will add in the losses of the test equipment to your
final result.

There is another way that steps around those problems. By using
three voltmeters it is possible to get all the data one needs to
calculate both core loss & PF.
It is based on the Law of Cosines for all the whiz kids out there.
The test setup is quite simple & consists of the source, a current
sampling resistor & the Device Under Test (DUT), all is series. For
the resistor I connected a set of four 100R, 2 watt resistors in
series-parallel. The resulting combination measured 98R on three
different ohmmeters, so that is what I used for the calculations.
I'm confident that the inductance of the combo will be low since
during my days in the R&D lab I had built attenuators good to 2GHz
in 50R with resistors of similar physical dimensions.

The three voltmeters measure the source, the drop across the series
R & the resulting voltage at the DUT. For the test setup I put
together, the source V1 is measured by a Pico Technology ADC-216
Virtual Instrument driving a PC. It has 16-bit resolution & can
simultaneously provide several readings. In this case it was used
for the voltage measurement & to monitor the waveform. V2 & V3 were
measured by a pair of 29S DMM's. The ADC-216 probe is set for 10X so
we have a 10M load at the source. Each of the 29S DMM's look like 5M
while in the ACV measurement mode. I avoided the use of clip leads
within the series circuit since that might cause additional losses
(And errors) to be included in the test data.

For the curious the equations used are as follows-

Power = ( V1^2 - V2^2 - V3^2 ) / 2R
PF = ( V1^2 - V2^2 - V3^2) / ( 2*V2*V3 )

The following are the test readings & results. The first set shows
frequency as the variable, while holding the DUT voltage within a
close range around 5 volts. Notice the losses seem to come to a
minimum at 3 KHz, then rise again. The equivalent parasitic
resistance is shown headed as Rsec. That is an imaginary resister
that if connected to the secondary would dissipate the same power as
shown on that line. The column headed Rpri is a similar imaginary
resistor if connected to the primary. It is as though we had added
another part when the transformer is connected in a real amplifier.

The second set of figures are pretty much what I had expected. Most
of the magnetic materials used in transformers of this kind have a
lower slope as they cross the vertical on the BH plane. That means
the equivalent parasitic resistance at low levels is less. In these
tests it appears we are nowhere near saturation. The last line would
deliver more than 32 watts to an 8 ohm load.

Doug Bannard has mentioned skin effect may be a problem as it would
increase copper losses & result in what would appear as additional
iron loss in these measurements as frequency is increased. I'm not
sure of the wire diameter here so don't know. Some may associate
skin effect with RF & MW frequencies. Turns out it can be used as
part of the starting winding in large synchronous motors. The
conductors on the rotor could be six inches deep. The rotating field
at start causes current to flow only in the upper part of the rotor
conductors, so starting current is held under control. I recall one
of the sub-station transformers I worked on at Ferranti had leads
brought out that were 6 inches wide by 3/8 inch thick copper. I
think it was 4000 amps per phase. The conductor had to be flat in
order to overcome skin effect.

No question skin effect is at work in the iron laminations & that is
why iron loss should go down as frequency is increased & lam
thickness is reduced. Not sure of the lam thickness here as well.
Hammond does use M6, just as many others do these days. It has
occurred to me that another kind of loss may be showing up here. All
of the insulation is a dielectric, which could also contribute to
the losses as frequency is increased.

BTW Doug, are you still at NT?


Frequency V1 V2 V3 Loss Rsec
Rpri PF
Hz pico 29S 29S mWatts

Meter Check 4.933 4.923 4.917 na na
na na

50 Hz 1.5789 11.757 5.011 438.5 57.26R
30.78K 0.73
100 13.573 9.512 5.039 348.8 72.80
39.13 0.71
300 10.579 6.321 5.003 239.4 104.55
56.20 0.74
1000 8.457 3.454 5.067 173.0 148.41
79.77 0.97
3 KHz 8.358 4.036 5.015 145.0 173.45
93.23 0.70
5 10.617 7.639 5.025 148.6 169.92
91.33 0.38
7 13.746 11.359 5.014 177.5 141.63
76.13 0.31
10 20.025 17.766 5.099 302.9 85.84
46.14 0.33
12 24.416 22.100 5.048 419.6 60.73
32.64 0.37

One KHz 1.2598 0.8105 0.628 2.734 144.25R
77.54K 0.53
3.746 1.8452 2.1155 31.39 142.57
76.63 0.79
8.947 3.670 5.451 188.1 157.97
84.91 0.92
14.345 5.377 8.913 497 159.84
85.92 1.00
19.404 6.972 12.28 903.6 166.89
89.70 1.00
25.615 8.888 16.494 1557 174.73
93.92 1.00

I've probably missed something, but it is time to quit! Cheers to
all, John Stewart

This set of data is more believeable to me, but the fact that the PF has a
peak at 1 kHz suggests that there are some resonant effects in the vicinity.

What are you using as a signal source? If the output impedance of the signal
source has an imaginary component, how will that affect your test results?

I'm out of town this week, but I will fool around with this technique myself
over the weekend.

"John Stewart" wrote in message
...

Rather then debate this subject aimlessly on the web, I decided to spend
some time getting some actual results. My previous experience going back too
many years had me convinced that as frequency was pushed up the transformer
losses increased. There must have been some basis for that, so I better have
a look in some detail in order to satisfy my curiosity. Was it core losses
or was it something else?
All of these tests were made on a recent production Hammond 1650N. It is
spec'd for 60 watts audio in PP & has a 4300 ohm primary with 40 % UL taps.
The secondary is the usual Hammond practice of three windings brought out to
provide match to 4, 8 & 16 ohm loads. The 8 ohm tap uses all of these
windings while the 4 & 16 ohm connexions use only 87 % of the secondary
copper. The impedance ratio, primary to secondary is 537.5. I used that
figure for the calculation of the primary parasitic resistance (Rpri in the
table below).
I had previously posted similar sets of data measured from a 50 year old
Hammond 1700 Series OPT & a recent production Hammond 125E Universal OPT.
Some of the posts from others got me suspicious of my test setup. I was
using a Metra-Hit 29S Precision DMM which has the ability to measure power,
both real & reactive & Power Factor (PF) along with all the other data. The
RMS responding VM in the 29S DMM is spec'd to 100 KHz. However, for power
the published spec goes only to one KHz. I had tried using it to 10 KHz &
got some results that were at least believable. As anyone who has ever tried
knows, the problem of making an AC current shunt of low resistance & large
bandwidth is daunting to say the least. Skin Effect is the culprit. That is
why you will not normally find AC current measurement to any great frequency
in a lower cost DMM. It is also the reason why the power measurement in the
29S is spec'd only to one KHz. Another consideration is 'how should the
connexion be made?' If one is not careful you will add in the losses of the
test equipment to your final result.
There is another way that steps around those problems. By using three
voltmeters it is possible to get all the data one needs to calculate both
core loss & PF.
It is based on the Law of Cosines for all the whiz kids out there. The test
setup is quite simple & consists of the source, a current sampling resistor
& the Device Under Test (DUT), all is series. For the resistor I connected a
set of four 100R, 2 watt resistors in series-parallel. The resulting
combination measured 98R on three different ohmmeters, so that is what I
used for the calculations. I'm confident that the inductance of the combo
will be low since during my days in the R&D lab I had built attenuators good
to 2GHz in 50R with resistors of similar physical dimensions.
The three voltmeters measure the source, the drop across the series R & the
resulting voltage at the DUT. For the test setup I put together, the source
V1 is measured by a Pico Technology ADC-216 Virtual Instrument driving a PC.
It has 16-bit resolution & can simultaneously provide several readings. In
this case it was used for the voltage measurement & to monitor the waveform.
V2 & V3 were measured by a pair of 29S DMM's. The ADC-216 probe is set for
10X so we have a 10M load at the source. Each of the 29S DMM's look like 5M
while in the ACV measurement mode. I avoided the use of clip leads within
the series circuit since that might cause additional losses (And errors) to
be included in the test data.
For the curious the equations used are as follows-
Power = ( V1^2 - V2^2 - V3^2 ) / 2R
PF = ( V1^2 - V2^2 - V3^2) / ( 2*V2*V3 )
The following are the test readings & results. The first set shows frequency
as the variable, while holding the DUT voltage within a close range around 5
volts. Notice the losses seem to come to a minimum at 3 KHz, then rise
again. The equivalent parasitic resistance is shown headed as Rsec. That is
an imaginary resister that if connected to the secondary would dissipate the
same power as shown on that line. The column headed Rpri is a similar
imaginary resistor if connected to the primary. It is as though we had added
another part when the transformer is connected in a real amplifier.
The second set of figures are pretty much what I had expected. Most of the
magnetic materials used in transformers of this kind have a lower slope as
they cross the vertical on the BH plane. That means the equivalent parasitic
resistance at low levels is less. In these tests it appears we are nowhere
near saturation. The last line would deliver more than 32 watts to an 8 ohm
load.
Doug Bannard has mentioned skin effect may be a problem as it would increase
copper losses & result in what would appear as additional iron loss in these
measurements as frequency is increased. I'm not sure of the wire diameter
here so don't know. Some may associate skin effect with RF & MW frequencies.
Turns out it can be used as part of the starting winding in large
synchronous motors. The conductors on the rotor could be six inches deep.
The rotating field at start causes current to flow only in the upper part of
the rotor conductors, so starting current is held under control. I recall
one of the sub-station transformers I worked on at Ferranti had leads
brought out that were 6 inches wide by 3/8 inch thick copper. I think it was
4000 amps per phase. The conductor had to be flat in order to overcome skin
effect.
No question skin effect is at work in the iron laminations & that is why
iron loss should go down as frequency is increased & lam thickness is
reduced. Not sure of the lam thickness here as well. Hammond does use M6,
just as many others do these days. It has occurred to me that another kind
of loss may be showing up here. All of the insulation is a dielectric, which
could also contribute to the losses as frequency is increased.
BTW Doug, are you still at NT?

Frequency V1 V2 V3 Loss Rsec Rpri PF
Hz pico 29S 29S mWatts
Meter Check 4.933 4.923 4.917 na na na na
50 Hz 1.5789 11.757 5.011 438.5 57.26R 30.78K 0.73
100 13.573 9.512 5.039 348.8 72.80 39.13 0.71
300 10.579 6.321 5.003 239.4 104.55 56.20 0.74
1000 8.457 3.454 5.067 173.0 148.41 79.77 0.97
3 KHz 8.358 4.036 5.015 145.0 173.45 93.23 0.70
5 10.617 7.639 5.025 148.6 169.92 91.33 0.38
7 13.746 11.359 5.014 177.5 141.63 76.13 0.31
10 20.025 17.766 5.099 302.9 85.84 46.14 0.33
12 24.416 22.100 5.048 419.6 60.73 32.64 0.37
One KHz 1.2598 0.8105 0.628 2.734 144.25R 77.54K 0.53
3.746 1.8452 2.1155 31.39 142.57 76.63 0.79
8.947 3.670 5.451 188.1 157.97 84.91 0.92
14.345 5.377 8.913 497 159.84 85.92 1.00
19.404 6.972 12.28 903.6 166.89 89.70 1.00
25.615 8.888 16.494 1557 174.73 93.92 1.00
I've probably missed something, but it is time to quit! Cheers to all, John
Stewart

BFoelsch wrote:

This set of data is more believeable to me, but the fact that the PF has a
peak at 1 kHz suggests that there are some resonant effects in the vicinity.

To the uninitiated, the use of a current source rather than a voltage source
to the primary of most *unloaded* OPTs will show that the primary input
impedance
looks like a low Q damped parallel LC circuit.
That is because we have Lp and Cshunt in parallel.
this may be 100H and 1,000 pF in parallel


But of course with iron wound things nothing is simple as we'd like, since the
actual Lp
isn't constant with F, since the U of the iron falls as F rises past an F where
the U for the iron has peaked.
Its not of any great concern if Lp is high enough, and Cshunt is low enough.
U at 1 kHz might be 1,000, not 5,000, like it might be at 50 Hz.

ZLp = L x F x 2 x 22/7.
So for 100H at 50 Hz, ZLp = 31,428 ohms.
L is proportional to U of the iron at the F concerned.

If the iron U was 1/5 the 50Hz value at 1 kHz, ( same voltage )
Lp at 1 kHz would be 20 x 1,000 x 6.28 = 12,560 ohms.
Cshunt of 1,000 pF = 159 kohms; it will have little effect at 1 kHz,
but at 10 kHz the C = 15.9 kohms.

If Lp = 20H, and Cshunt was 1,000 pF, the resonant F = 251 Hz.
So with a current source to the input of an OPT, the peak in Za-a with no load
may not be at 1 kHz, and the response will be anything but flat.
And the peak in the response will vary with voltage applied. Its queer.
Using a low R input source, say triodes, and terminating the secondary
with some certain R values will flatten the response, make the core and
capacitance losses
insignificant, and make the load value change minimal for the most of the audio
band.

If RLa-a = 3k as it may where LP was 100H at normal levels, the change in ZLp
may have little effect on the load value across the audio band.

Good OPTs have very fine thin laminations of GOSS with a high U,
and HF losses are minimised and the design seeks to minimise
Cshunt especially where RL is a highish figure like 10k a-a.

I had a guy visit me with an amp with a quad of KT88 in triode where the
Cshunt in the 1.6k : 8 ohm OPT was so high that the HF pole was at 8 kHz
with the amp loaded with 8 ohms, open loop response.

There were about 6 secondary sections interleaved with 7 primary sections,
and leakage inductance was less than a mH, very good, but too good,
because the P&S sectiions were wound with very thin PS insulation of only
0.1 mm perhaps, and all the interleavings raised Cshunt to a very high figure
and mauled the HF performance.
The guy who wound this nightmare tranny had good intentions, "couple it tightly"

he must have thought, but only to the detriment of the audio outcome.

RDH4 gives info and a formula on how to address the issue of optimal numbers of
interleavings for a
given transformer.

P to S insulation above 0.5 mm is to be preferred for **all** OPTs certainly
because
of the Cshunt concerns, but also because of voltage insulation requirements.

If lots of interleavings are to be used, the insulation thickness needs to be
increased, not decreased
to reduce C-shunt.
LL will be still low even if the distance between P&S sections is raised to say
1 mm
in a typical 100 watt OPT with 4P and 5S sections.
The LL reduces nearly proportionally to the square of the number of
interleavings, so its the no of sections that reduces LL far more than the
distance
between P and S sections.
Thus a large window area is to be preferred, to fit in all those layers of wire,

and insulation, and still achieve low winding losses.

The windup details of the Williamson OPT given in RDH4
include the insulation details, and the thickness of the "empire tape"
between P&S sections wasn't just a guess; it was very critical to the success of
the design
for wide BW and low losses, and not just core losses as opposed to LP and Cshunt
losses.
Williamson specified a special non wasteless pattern of E&I core material;
the tongue width is 32 mm, but the window is 75 x 25 mm.

Conrad Johnson do the same trick to get the most from a given weight of iron
in their 140 watter I examined last week during servicing.
A company like that or probably their sub contractor has a mill to cut special
size lams
nobody will see often.

Leak also did the same.
Such specially cut laminations are not easy to find, so another solution is to
use c-cores because there
are plenty of C-cores with a smallish centre leg area but with a large widow
size.

I actually prefer to use wasteless core pattern reduce the turn count, and use a
high stack of iron.
I normally use 50% more tranny weight than the more mass market commercial
brands.

I am forced to refer to RDH4 Williamson OPT recipe, because
SFA other tranny makers **ever** disclose the exact similar specifications
for their wares. They hold their cards very close their chests indeed.
The specification they offer is marketing for ignorant DIYers, rather than
information which an
EE could really check right out.
There are, however, full details of OPT No1 at my website.

Patrick Turner.


















What are you using as a signal source? If the output impedance of the signal
source has an imaginary component, how will that affect your test results?

I'm out of town this week, but I will fool around with this technique myself
over the weekend.

"John Stewart" wrote in message
...

Rather then debate this subject aimlessly on the web, I decided to spend
some time getting some actual results. My previous experience going back too
many years had me convinced that as frequency was pushed up the transformer
losses increased. There must have been some basis for that, so I better have
a look in some detail in order to satisfy my curiosity. Was it core losses
or was it something else?
All of these tests were made on a recent production Hammond 1650N. It is
spec'd for 60 watts audio in PP & has a 4300 ohm primary with 40 % UL taps.
The secondary is the usual Hammond practice of three windings brought out to
provide match to 4, 8 & 16 ohm loads. The 8 ohm tap uses all of these
windings while the 4 & 16 ohm connexions use only 87 % of the secondary
copper. The impedance ratio, primary to secondary is 537.5. I used that
figure for the calculation of the primary parasitic resistance (Rpri in the
table below).
I had previously posted similar sets of data measured from a 50 year old
Hammond 1700 Series OPT & a recent production Hammond 125E Universal OPT.
Some of the posts from others got me suspicious of my test setup. I was
using a Metra-Hit 29S Precision DMM which has the ability to measure power,
both real & reactive & Power Factor (PF) along with all the other data. The
RMS responding VM in the 29S DMM is spec'd to 100 KHz. However, for power
the published spec goes only to one KHz. I had tried using it to 10 KHz &
got some results that were at least believable. As anyone who has ever tried
knows, the problem of making an AC current shunt of low resistance & large
bandwidth is daunting to say the least. Skin Effect is the culprit. That is
why you will not normally find AC current measurement to any great frequency
in a lower cost DMM. It is also the reason why the power measurement in the
29S is spec'd only to one KHz. Another consideration is 'how should the
connexion be made?' If one is not careful you will add in the losses of the
test equipment to your final result.
There is another way that steps around those problems. By using three
voltmeters it is possible to get all the data one needs to calculate both
core loss & PF.
It is based on the Law of Cosines for all the whiz kids out there. The test
setup is quite simple & consists of the source, a current sampling resistor
& the Device Under Test (DUT), all is series. For the resistor I connected a
set of four 100R, 2 watt resistors in series-parallel. The resulting
combination measured 98R on three different ohmmeters, so that is what I
used for the calculations. I'm confident that the inductance of the combo
will be low since during my days in the R&D lab I had built attenuators good
to 2GHz in 50R with resistors of similar physical dimensions.
The three voltmeters measure the source, the drop across the series R & the
resulting voltage at the DUT. For the test setup I put together, the source
V1 is measured by a Pico Technology ADC-216 Virtual Instrument driving a PC.
It has 16-bit resolution & can simultaneously provide several readings. In
this case it was used for the voltage measurement & to monitor the waveform.
V2 & V3 were measured by a pair of 29S DMM's. The ADC-216 probe is set for
10X so we have a 10M load at the source. Each of the 29S DMM's look like 5M
while in the ACV measurement mode. I avoided the use of clip leads within
the series circuit since that might cause additional losses (And errors) to
be included in the test data.
For the curious the equations used are as follows-
Power = ( V1^2 - V2^2 - V3^2 ) / 2R
PF = ( V1^2 - V2^2 - V3^2) / ( 2*V2*V3 )
The following are the test readings & results. The first set shows frequency
as the variable, while holding the DUT voltage within a close range around 5
volts. Notice the losses seem to come to a minimum at 3 KHz, then rise
again. The equivalent parasitic resistance is shown headed as Rsec. That is
an imaginary resister that if connected to the secondary would dissipate the
same power as shown on that line. The column headed Rpri is a similar
imaginary resistor if connected to the primary. It is as though we had added
another part when the transformer is connected in a real amplifier.
The second set of figures are pretty much what I had expected. Most of the
magnetic materials used in transformers of this kind have a lower slope as
they cross the vertical on the BH plane. That means the equivalent parasitic
resistance at low levels is less. In these tests it appears we are nowhere
near saturation. The last line would deliver more than 32 watts to an 8 ohm
load.
Doug Bannard has mentioned skin effect may be a problem as it would increase
copper losses & result in what would appear as additional iron loss in these
measurements as frequency is increased. I'm not sure of the wire diameter
here so don't know. Some may associate skin effect with RF & MW frequencies.
Turns out it can be used as part of the starting winding in large
synchronous motors. The conductors on the rotor could be six inches deep.
The rotating field at start causes current to flow only in the upper part of
the rotor conductors, so starting current is held under control. I recall
one of the sub-station transformers I worked on at Ferranti had leads
brought out that were 6 inches wide by 3/8 inch thick copper. I think it was
4000 amps per phase. The conductor had to be flat in order to overcome skin
effect.
No question skin effect is at work in the iron laminations & that is why
iron loss should go down as frequency is increased & lam thickness is
reduced. Not sure of the lam thickness here as well. Hammond does use M6,
just as many others do these days. It has occurred to me that another kind
of loss may be showing up here. All of the insulation is a dielectric, which
could also contribute to the losses as frequency is increased.
BTW Doug, are you still at NT?

Frequency V1 V2 V3 Loss Rsec Rpri PF
Hz pico 29S 29S mWatts
Meter Check 4.933 4.923 4.917 na na na na
50 Hz 1.5789 11.757 5.011 438.5 57.26R 30.78K 0.73
100 13.573 9.512 5.039 348.8 72.80 39.13 0.71
300 10.579 6.321 5.003 239.4 104.55 56.20 0.74
1000 8.457 3.454 5.067 173.0 148.41 79.77 0.97
3 KHz 8.358 4.036 5.015 145.0 173.45 93.23 0.70
5 10.617 7.639 5.025 148.6 169.92 91.33 0.38
7 13.746 11.359 5.014 177.5 141.63 76.13 0.31
10 20.025 17.766 5.099 302.9 85.84 46.14 0.33
12 24.416 22.100 5.048 419.6 60.73 32.64 0.37
One KHz 1.2598 0.8105 0.628 2.734 144.25R 77.54K 0.53
3.746 1.8452 2.1155 31.39 142.57 76.63 0.79
8.947 3.670 5.451 188.1 157.97 84.91 0.92
14.345 5.377 8.913 497 159.84 85.92 1.00
19.404 6.972 12.28 903.6 166.89 89.70 1.00
25.615 8.888 16.494 1557 174.73 93.92 1.00
I've probably missed something, but it is time to quit! Cheers to all, John
Stewart

There is a simpler way --

Get a clean paint can, fill with vegetable oil. Immerse a 10 ohm 20 watt resistor in the oil and flow one amp through the resistor. Track temperature and time, gently stir the oil during this period. When it plateus you have a factor which will allow you to calculate the heat transfer coefficient of the salad oil.

You can dunk the transformer in the oil by itself or enclose it in a baggie and place in the oil. The volume of oil should be the same as that used when the coefficient measurement was performed. Place the appropriate load on the trannie. Measure the temperature rise in the oil over a comparable unit of time, and voila you have the loss due to heating effects, wire resistance etc.

In essence, this is how a Bird Wattmeter, or HP-Agilent Power Meter really works. You don't have to worry about the trig.



"John Stewart" wrote in message ...

Rather then debate this subject aimlessly on the web, I decided to spend some time getting some actual results. My previous experience going back too many years had me convinced that as frequency was pushed up the transformer losses increased. There must have been some basis for that, so I better have a look in some detail in order to satisfy my curiosity. Was it core losses or was it something else?
All of these tests were made on a recent production Hammond 1650N. It is spec'd for 60 watts audio in PP & has a 4300 ohm primary with 40 % UL taps. The secondary is the usual Hammond practice of three windings brought out to provide match to 4, 8 & 16 ohm loads. The 8 ohm tap uses all of these windings while the 4 & 16 ohm connexions use only 87 % of the secondary copper. The impedance ratio, primary to secondary is 537.5. I used that figure for the calculation of the primary parasitic resistance (Rpri in the table below).

I had previously posted similar sets of data measured from a 50 year old Hammond 1700 Series OPT & a recent production Hammond 125E Universal OPT. Some of the posts from others got me suspicious of my test setup. I was using a Metra-Hit 29S Precision DMM which has the ability to measure power, both real & reactive & Power Factor (PF) along with all the other data. The RMS responding VM in the 29S DMM is spec'd to 100 KHz. However, for power the published spec goes only to one KHz. I had tried using it to 10 KHz & got some results that were at least believable. As anyone who has ever tried knows, the problem of making an AC current shunt of low resistance & large bandwidth is daunting to say the least. Skin Effect is the culprit. That is why you will not normally find AC current measurement to any great frequency in a lower cost DMM. It is also the reason why the power measurement in the 29S is spec'd only to one KHz. Another consideration is 'how should the connexion be made?' If one is not careful you will add in the losses of the test equipment to your final result.

There is another way that steps around those problems. By using three voltmeters it is possible to get all the data one needs to calculate both core loss & PF.
It is based on the Law of Cosines for all the whiz kids out there. The test setup is quite simple & consists of the source, a current sampling resistor & the Device Under Test (DUT), all is series. For the resistor I connected a set of four 100R, 2 watt resistors in series-parallel. The resulting combination measured 98R on three different ohmmeters, so that is what I used for the calculations. I'm confident that the inductance of the combo will be low since during my days in the R&D lab I had built attenuators good to 2GHz in 50R with resistors of similar physical dimensions.

The three voltmeters measure the source, the drop across the series R & the resulting voltage at the DUT. For the test setup I put together, the source V1 is measured by a Pico Technology ADC-216 Virtual Instrument driving a PC. It has 16-bit resolution & can simultaneously provide several readings. In this case it was used for the voltage measurement & to monitor the waveform. V2 & V3 were measured by a pair of 29S DMM's. The ADC-216 probe is set for 10X so we have a 10M load at the source. Each of the 29S DMM's look like 5M while in the ACV measurement mode. I avoided the use of clip leads within the series circuit since that might cause additional losses (And errors) to be included in the test data.

For the curious the equations used are as follows-

Power = ( V1^2 - V2^2 - V3^2 ) / 2R
PF = ( V1^2 - V2^2 - V3^2) / ( 2*V2*V3 )

The following are the test readings & results. The first set shows frequency as the variable, while holding the DUT voltage within a close range around 5 volts. Notice the losses seem to come to a minimum at 3 KHz, then rise again. The equivalent parasitic resistance is shown headed as Rsec. That is an imaginary resister that if connected to the secondary would dissipate the same power as shown on that line. The column headed Rpri is a similar imaginary resistor if connected to the primary. It is as though we had added another part when the transformer is connected in a real amplifier.

The second set of figures are pretty much what I had expected. Most of the magnetic materials used in transformers of this kind have a lower slope as they cross the vertical on the BH plane. That means the equivalent parasitic resistance at low levels is less. In these tests it appears we are nowhere near saturation. The last line would deliver more than 32 watts to an 8 ohm load.

Doug Bannard has mentioned skin effect may be a problem as it would increase copper losses & result in what would appear as additional iron loss in these measurements as frequency is increased. I'm not sure of the wire diameter here so don't know. Some may associate skin effect with RF & MW frequencies. Turns out it can be used as part of the starting winding in large synchronous motors. The conductors on the rotor could be six inches deep. The rotating field at start causes current to flow only in the upper part of the rotor conductors, so starting current is held under control. I recall one of the sub-station transformers I worked on at Ferranti had leads brought out that were 6 inches wide by 3/8 inch thick copper. I think it was 4000 amps per phase. The conductor had to be flat in order to overcome skin effect.

No question skin effect is at work in the iron laminations & that is why iron loss should go down as frequency is increased & lam thickness is reduced. Not sure of the lam thickness here as well. Hammond does use M6, just as many others do these days. It has occurred to me that another kind of loss may be showing up here. All of the insulation is a dielectric, which could also contribute to the losses as frequency is increased.

BTW Doug, are you still at NT?


Frequency V1 V2 V3 Loss Rsec Rpri PF
Hz pico 29S 29S mWatts

Meter Check 4.933 4.923 4.917 na na na na

50 Hz 1.5789 11.757 5.011 438.5 57.26R 30.78K 0.73
100 13.573 9.512 5.039 348.8 72.80 39.13 0.71
300 10.579 6.321 5.003 239.4 104.55 56.20 0.74
1000 8.457 3.454 5.067 173.0 148.41 79.77 0.97
3 KHz 8.358 4.036 5.015 145.0 173.45 93.23 0.70
5 10.617 7.639 5.025 148.6 169.92 91.33 0.38
7 13.746 11.359 5.014 177.5 141.63 76.13 0.31
10 20.025 17.766 5.099 302.9 85.84 46.14 0.33
12 24.416 22.100 5.048 419.6 60.73 32.64 0.37

One KHz 1.2598 0.8105 0.628 2.734 144.25R 77.54K 0.53
3.746 1.8452 2.1155 31.39 142.57 76.63 0.79
8.947 3.670 5.451 188.1 157.97 84.91 0.92
14.345 5.377 8.913 497 159.84 85.92 1.00
19.404 6.972 12.28 903.6 166.89 89.70 1.00
25.615 8.888 16.494 1557 174.73 93.92 1.00

I've probably missed something, but it is time to quit! Cheers to all, John Stewart

"John Stewart" wrote

snip...can't repro table in plain text

There appears to be an anomaly in your measurements that calls into
question their claimed precision.

Compare the results for 1k @ 5.067V in your first block of data,
with those for 1k @ 5.451V in the second block. For example, the PF
is shown as 0.97 in the former, and 0.92 in the latter. Looking at
the trend in the second block, the former should be the smaller
value. Quoting to 2 significant figures seems unjustified. Could be
temperature difference between the two blocks I suppose, so
comparison within blocks may still be worth the digits.

On the question of where the losses go, several contributions have
mentioned the primary capacitance, Cp. Assuming your measurement
method is correct and you have isolated the power consumed by the
transformer, the current arising from Cp will not directly result in
power loss unless it is in series with the primary winding
resistance. There is also the leakage inductance to take into
account. Losses in the secondary due to Cp are presumably included
in your measurements.

Even if you assume that the entire primary winding resistance is in
series with Cp, the loss seems to my mental arithmetic to be an
order of magnitude or two short of explaining much.

The question of whether current due to Cp passes through the winding
resistance is a problem for simple simulation models, where
different configurations are often used for low and high
frequencies.

One way round the problem is to model using a perfect transformer of
infinite inductance, in parallel with a network to model losses.
Best I have found are at:

http://www10.pcbcafe.com/book/parse_...F232coresb.htm

which omits capacitance altogether, and

http://www.beigebag.com/case_xfrmer_2.htm

Significantly not all include core losses, and none take into
account variation in winding resistances due to skin effect or
whatever. But it would be interesting to compare simulated losses
with your real data, then wonder about the differences.

Unfortunately the code includes a fair amount of Pspice and other
stuff that my version won't accept, so it will take me a while to
translate. Other ppl here may be able to run your experiment in a
jiffy.

cheers, Ian

BFoelsch wrote:

This set of data is more believeable to me, but the fact that the PF has a
peak at 1 kHz suggests that there are some resonant effects in the vicinity.

What are you using as a signal source?

It is a 35 watt tubed amp that doesn't seem to mind reactive loads at all.

If the output impedance of the signal
source has an imaginary component, how will that affect your test results?

You have a point & I have to say I was somewhat concerned of that as well.
Some time ago I did some reverse tests where I drove one amp from another
to test for the internal Z of one of them. Again, I had used one of the 29S
DMM's
to keep track of things & found the internal Z of the amp in test went from
inductive
at some low frequency to capacitive at higher frequencies. Don't recall at what
f but
seems it was somewhere in the mid-range, just as you would expect.

Later, using the 3-voltmeter method of power measurement I think the source
reactance doesn't matter much. There is nowhere in the equations or mention of
that in the very thick text book I pulled it out from.

See the many comments by others. That should shed some light on what is
happening.

Cheers, John Stewart



I'm out of town this week, but I will fool around with this technique myself
over the weekend.