Unless loudspeakers are tested and measured adequately during their

manufacture or repair work, there is NO CHANCE that the result will

ever sound very good. So the first item for this page on DIY speakers

manufacture deals with testing, because its most important, and nobody

can ever fully trust their ears.

Many people use an enormous amount of time to make wooden boxes for
"normal" front radiating dynamic drive units and some even make complex
horn structures for horn loaded drivers hidden from view, and all without
real understanding of the electronics or maths involved or any use of adequate
testing procedures, so they end up constructing what I call noisy expensive
firewood. But at least good measuring gear allows anyone to see their own
shortcomings, if they are not the type who thinks he cannot ever make a
mistake or cannot be any better.

I built my own very adequate test equipment to analyze what is right or wrong
with the F response of any speaker. It is all analog and rather old fashioned.

There is perhaps no doubt that modern PC based programs used with a sound
card with built in pink noise source and a decent calibrated microphone will
enable even fairly inexperienced diyers to measure their own speakers.
I have not found I have any need to use a fully computer aided test method.
I would have to set up a PC in my listening room and have yet another
table full of clunky dust gathering junk always in my way when I am not
testing speakers. Those with laptops will cope better.

The computer program approach has major advantages. The filter can have
a higher number of bands, filtering is all done digitally, so Q could be higher,
and measurement is achieved much faster, because what I do manually is done
automatically. Just whether its more accurate and useful is a moot point.
There were plenty of expensive and well manufactured speakers made before
any computers were used during design and manufacture. There are now very
many very poor quality speakers made even though manufacturers have access
to very cheap computing power.
I guess if I did have a PC with sound card and an analysis program, I could check
it against my old fashioned test gear I show here, and then see which seemed to
be telling the truth or not.

Fig 1. Speaker testing set up.

The above diagram shows a power amp, microphone and a typical speaker arranged
for response testing. The pink noise test signal, microphone amp, and 34 band filter
is in one metal box about the size of a large shoe box and powered from a mains
transformer and rectifier giving + / - 14Vdc supply rails for the op-amps, transistors,
diodes, switches, analog meter, and active and passive filter circuits with R&C elements.
There are no inductors used anywhere.

Fig 2. Pink Noise Generator.

Electronic "white" noise is generated using a single transistor Q1 with a low amount of
current flow in the backward or reverse direction to normal operation. The white-noise
from such a simple device is a small signal containing randomly varying frequencies,
with randomly varying amplitudes, phase, and the amplitudes rise at 3dB/octave
over the audio band and beyond, and limited in rise at RF because of stray capacitance.
An op-amp U1 amplifies the noise. The op-amp has NFB and attenuation of HF above
25kHz. There is a following R&C filter network giving attenuation of -3dB per octave
so that the peak amplitude remains constant to 20kHz, and thus white-noise is converted
to pink-noise. A second op-amp U2 amplifies the noise slightly, and its following R&C
network shapes the noise amplitude contour to be flat because the initial transistor does
not produce an ideal noise signal with equal energy in an equal spaced range of F.
U3 is a unity gain buffer to power a 20k log pot allowing level of pink noise to be adjusted.

Bandwidth of the pink noise is 16Hz to 23kHz, -3dB points. Such a noise signal is excellent
for testing loudspeakers.

The equalizing circuit is extremely important so that when any band is selected using a
bandpass filter with 36 switchable bands, the average amplitude is the same for each band,
where the Q of each band is between say 5 and 12. I have not tried having a BPF with
Q of say 24, and with many more switchable bands. Having 36 bands is probably enough
to capture the average level of each band. In any speaker, it would be unusual for a peak
or valley in the acoustic response to have a Q of more than 5. For example, say there was
a sharp peak in response of +6dB at say 47Hz, then the readings of energy in bands
at 43Hz and 52Hz would both be pushed up about +3dB. Very sharp response peaks
or valleys cannot usually be compensated by LCR crossover networks, but they can't be
ignored either, but arranging the crossover to have a -3dB cut in the 2 bands of 43Hz and
52 Hz is feasible, and will lessen the effect of a peak.  Having a 72 band BPF may just
drive the speaker designer mad, because he is made aware of problems he may not
be able to cure.

A suitable bandpass filter is best explained to you all by starting with basics.

Fig 3, response of bandpass filter ( BPF ) and the basic schematic :-

Fig 3 shows a basic active BPF schematic on LHS which uses a "bridged T R&C notch filter"
for a negative feedback network applied between the low Ro output of an op-amp to the high
Rin inverting terminal of an op-amp. At very low F or very high F the op-amp voltage gain = 1.
But at Fo, op-amp gain becomes high because the NFB signal applied is much lower
than at op-amp Vo, because the filter network produces a null in the response at Fo.
The op-amp input circuit shows VR1 to set input level from any noise signal
source such as an amplified microphone signal picking up sound from a speaker driven with
pink-noise. C1 & R1 and R2 form a HPF to keep extremely low F out of the op-amp, with
R2 biasing the non-inverting op-amp input at 0V. R3 and R4 form a positive FB network
which about doubles the gain of the op-amp and increases the Q of the filter.
The bridged T RC filter has two R labelled Rb, and these are equal value R. But
each can be changed in value to alter the Fo to a desired F.

The actual BPF I have used has three selectable F ranges with each having 12 selectable
bands, so there are a total of 36 bands which can be tested......... 

Fig 4. 36 band BPF schematic.

Fig 4 shows the bandpass filter consisting of a "bridged 'T' RC filter" in a series NFB loop
around an op-amp. This gives Q = 5 for each band of F. Adding about 6dB of positive FB via
R3 18k0 and R4 2M0 increases Q to about 12, and although there are "gaps" between pass bands,
the F response recorded seemed better than if the PFB had been left out and Q of 5 used.  

The Q is calculated as Q = Center F / ( F2 - F1 ) where F1 is the LF -3dB cut off,
and F2 is the HF -3dB cut off.
For example, Fig 3 shows a response for F0 = 52Hz. F1 = 47Hz and F2 =57Hz,
so the bandwidth = 57 - 47 = 10Hz, and Q = 52 / 10 = 5 approximately.

It is unusual that peaks and dips in the speaker response will have sharper Q or higher
rates of attenuation in speaker speaker than any one filter pass band, so usually the
truth about a response can be measured using the filter as I have it.

Fig 4 shows a bandpass filter with 36 switchable bands from 24Hz to 20kHz.
Bass band, Hz, :- 24, 29, 35, 43, 52, 63, 76, 92, 112, 136, 165, 200.
Midrange band, Hz, :- 240, 290, 350, 430, 520, 630, 760, 920, 1,120, 1,360, 1650, 2,000.
Treble band, kHz, :- 2k4, 2k9, 3k5, 4k3, 5k2, 6k3, 7k6, 9k2, 11k2, 13k6, 16k5, 20k0.

To achieve the centre F of each pass band, there are two series strings of resistors
soldered around a 12 position 2 pole wafer switch, which was new old stock made in about 1950,
and still very reliable after 12 years of use.

The R values were carefully worked out to give bands as listed above, and accurate within
+/-3% of the nominal F. To achieve all the values of R11 to R34, you should use parallel and/or
series connected arrangements of 1% x 1/4 Watt metal film resistors which are quite
cheaply purchased in resistor packs containing say 10 R of all standard values between
10ohms and 1Meg-ohm.
Accuracy of the final values should be within +/- 3% as calculated. Don't rely on trials of
parallel R using an ohm meter, so you need to be able to calculate or access a calculator
program which tells you what R to use to make up each of the values needed above. 

For example, to make up R11 = 1,695 ohms,
Go to http://www.sengpielaudio.com/calculator-paralresist.htm 
There is a table down the page where you can choose two standard resistance values
to get close to the wanted total.
Example :- Find 2 parallel R to give slightly under 1,695 ohms, = 2k2 // 6k8 = 1,662 ohms.

Then, 1695 ohms - 1662 ohms = 33 ohms.
So you could use ( 2k2 // 6k8 ) + 33 ohms = 1,695 ohms total. 

The Resistance Calculator allows entering the total wanted R plus some other value R1 above the
wanted value, and then the needed R2 parallel value is calculated.
So you get 2k2 // 7k384 to make 1,695 ohms.
But 7k384 is not near a standard value except 7k5 which isn't included in resistor packs, usually
just 5k6 and 6k8. So try final value = 7k384, and let R3 = 10k0, and it calculates
R4 = 28k, and final R = 1,694 ohms. Now 27k is a standard value closest to 28k, and if you use
2k2 // 10k0 // 27k0, you get 1,690 ohms, and error is 0.3% and quite OK.
Using only paralleled R means their leads may be all twisted up and soldered with only one lead
at each end left long so the compound R is easy to fit and solder neatly around the switch lugs.

There are THREE frequency ranges, with each determined by a switched pair of capacitors
using a 2 pole x 3 position wafer switch. Some sort of press button switch could be used for changing
cap values. 

The capacitor values are :-
Bass 4.7uF + 47nF, Midrange 0.47uF + 4n7, Treble 47nF + 470pF.
Good quality polypropylene caps should suffice, but to ensure the values are close to those
nominated, the C should be measured first. When the bandpass filter is completed, it must be
tested to ensure your work is done correctly. To check the capacitor values for each F range, use a
sine wave signal generator with variable F but fixed amplitude to apply signal to the lowest and
highest F able to be selected. The signal is varied to "tune" for highest amplitude of Fo.
A frequency meter must be use to record the frequency. If you get say 20Hz instead of the wanted 24Hz,
the C pillar of 4u7 must be made smaller, so use multiple smaller paralleled capacitors to achieve the
wanted F. If you get say 30Hz for the lowest Fo, you need to add parallel C to 4u7 until you have
24Hz. At the high end of the band, the 47n may need adjustment to get the wanted 200Hz.
Once that is done, you should get all the other Fo I have listed.

The bandpass profiles of all 36 bands and their centre frequency can be plotted and recorded.
You should be able to record the F1 and F2 for each band where the response is -3dB, ie,
if Vout = 1V at Fo, you need to record F where Vout = 0.7V.
What you should find :-
All Vout for 36 bands have the same amplitude at Fo, except for 24Hz and 20kHz, which may
be slightly down because of the pink noise pass band contouring.
Fo where maximum Vout occurs should be within +/- 5% of the bands listed above, and Q of all bands
should be equal. I have Q = 12 with some slight positive FB around the BPF opamp.
Without PFB, the Q should be about 5.0
A low Q below 4 or a high Q above 15 wouldn't be useful.

When your BPF passes all tests using sine waves, it should be used to test your pink-noise signal to
ensure an equal amount of energy exists in each F band. After applying pink-noise to the BPF,
the meter Vout should be equal for each band. The response for all 36 bands can be plotted as a
graph and the response should show some reduction in level at 24Hz and 20kHz, but for
all other F the response should be no more than +/- 1dB above or below a reference level at say 920Hz
It takes awhile to get used to levels needed to get above noise but below overload of any part
of the circuit. What you will find is that the meter amp as I have it will not give a rock steady voltage reading.
This is because the amplitudes of all bands of signals vary at a random rate which includes LF rates which
tend to get past the time constants of peak detector circuit and meter circuit. The time constants should
not be any slower, ie, C3 and C4 in Fig 4 should not be higher values because you need the meter
to be able to swing to a different amplitude at a different Fo without having to wait too long. So when reading
the meter, its normal to see the meter needle wobbling +/- 2dB around a centre value and you have to
estimate the reading.
THE METER should have a nice big meter face. Mine is 100mm wide, and maybe 40 years old. I removed the
glass face and attached a new cardboard blank dial to accept calibration markings with blank ink pen.
Setting up the meter takes some doing, because you have to figure out how much signal is needed to
swing the meter fully, and then you need to apply an adjusted fixed Vdc across the meter, maybe similar
to how I have mine with R8, R9, R10. The reason for the slight negative voltage at the negative meter terminal
is because you want the meter to read lowest reading -20dBV with say 0.01Vac signal, 0.1Vac at center reading,
0.0dBV, and 1Vac at the maximum reading, +20dBV. The dc op-amp U3 has a diode in a shunt NFB loop and it is
LOGARITHMIC amplifier. How I have it allows easy readings of voltages varying over a 100 fold value,
something quite impractical if the meter amp was a linear type. The meter face can be calibrated in 3dB steps,
and you will find all to be about equally spaced so thus can be divided up to give you even 1dB spacings.
CALIBRATING the meter is done using a sine wave applied to the peak detector input and while measuring
its Vac input value. The range of Vac produced at the input to the peak detector should be well above circuit noise
but below any overloading. Such matters take days and days to get correct.
Building my "box of tricks" took weeks of full time work with many late nights. If you only get 4 hours each
Saturday afternoon, expect to never ever finish any hand crafted electronics including any amplifiers or speakers.

If you were extremely keen, you could use a 2 pole x 24 position rotary switch to change
the resistance values. This means you could have 72 F bands, with 24 F in each band.
F band Hz numbers would then be :-
22, 24, 26, 29, 32, 35, 39, 43, 47, 52, 57, 63, 69, 76, 84, 92, 101, 112, 123, 136, 150, 165, 181, 200.
Q would need to be higher.
2 pole x 24position switches are available for constructing stereo switched attenuator volume controls.
But the switch needs to be very good quality to avoid intermittent contacts and perplexing variations
and annoyances during measurement.


High pass filter.
If you look closely, you can see a switchable high pass filter after the
microphone amp. This filters out large stray LF signals the microphone
may have picked up below 300Hz, and which may affect the measurement
of frequencies above 300Hz by causing excessive meter needle movement.
This can be a useful thing when trying to measure midrange and treble units
in an environment where there is extraneous LF noise such as traffic.
The whole unit I made is portable, and I have been asked to check systems
in homes of other people, and this is all usually just as easy as testing
speakers here.

Microphone and amp.
In about 1996, I made 3 microphones.
I use pre-tested and calibrated electret mic inserts from Hi-Q International in
Melbourne in my home built microphones. They cost less than $2 each.
Such inserts are about 10mm in dia and two wires can be soldered to the rear and
then the mic "button insert" is fitted to a 300mm long length of 12mm copper
water pipe.
The mic is padded in cotton wool where it is flush with the pipe end to prevent
hard contact with the pipe. An RCA socket is fitted to the far end of the pipe,
and the cost of such a good mic is $10 and some time. I use an old heavy stage
mic stand and clamp to alter mic height and position. I occasionally a change
mics to check on each other for drift, but after several years there has not been
much change and I continue to be able to build new speakers of reform older
speakers and have them sing very well.

Behringer electret microphone.
A friend provided me with a new Behringer Calibrated Microphone which cost
about $150 in 2007 and I tested a pair of speakers I was reforming at that
time using both my own hand made microphones and the swish looking
Behringer. Mics were set up 150mm apart and at the 3 metre distance from
the speakers in my shed which is so cluttered that it is fairly non resonant.
The response graphs plotted from meter readings and for 20Hz to 20kHz
varied less than +/- 1dB and after a moving mics a little but keeping them
close to each other, the variations when averaged became less than +/- 1dB.
I figured my homemade mics were just as good as the factory made
"professional" microphones.

Measuring acoustic signal levels.
I set up speakers as they may be used in most hi-fi listening situations in a
normal lounge room.
Small floor standers should be raised off the floor so the tweeter is about
1 metre above floor level. My loungeroom is alos cluttered, but carpeted and
with enough soft furnishings to damp many resonances. There are sloping
ceilings and an irregular wall layout and a volume all up of 200 cubic metres.
Most people very much like my lounge room's acoustic properties. If I
get a pair of speakers to sound well and measure well in my room, then they
usually "travel well" and sound very well in anyone else's home.

When pink noise is used for a test signal the rooms resonant behaviour is much
avoided. If pure steady F sine waves are used to test speakers in a normal room,
one will get a graph with perhaps 15 peaks and troughs of up to +/- 15dB across
the whole audio band and the measurement of speaker output for narrow F bands
is entirely ruined by the resonant behaviour. Anechoic chambers are needed
for sine wave response measurements, and such chambers or rooms are large,
and only available at huge expense. But by using pink noise which resembles
the sound of a huge nearby waterfall, the phase of any one frequency within
the noise varies so much that an outgoing wave at say 1kHz cannot much
excite the resonant waves which exist in a normal echoic room, ie, a normal
lounge room. The effects of the worst resonances of a room are limited to 3dB
variation in voltage measurements from the microphone.
The 4 mic positions are usually at 3.5M +/- 0.5M from a speaker and nearly on
axis and between 0.7M and 1.2M above floor level. The response will vary slightly
with each of the 4 mic positions because even with randomly varying test signal
frequency and phase shift and level there is still some effect from room
resonances and reflections.

I usually set up the pink noise generator level to give a healthy sound level
which is loud enough to drown out other noises in the environment.
SPL is roughly equal to an 88dB SPL reading from an SPL meter one might
buy. Having mentioned such meters, don't be tempted to use one for response
checks, they are too inaccurate, and you must apply signal to the speaker using
a filtered band of frequencies from the pink noise. I find it better to apply the
whole band of pink noise 20Hz to 20kHz, and then filter the recovered and
amplified microphone signal.
The best time to measure speakers is late at night when LF traffic rumble is
at a minimum, and you can turn off any noisy devices, and rumble from
passing traffic or lawn mowers is at a minimum.

The graphs you may wish to plot for speakers may be drawn on printed paper
copies of the blank graph sheet shown here :-

This blank sheet may be printed off by anyone wanting to plot F responses of speakers.

There are TWO frequency scales. One is arranged so there is even spacing of F bands
along the audio band but the rate of F increase is actually LOGARITHMIC, and there
is a log scale at the bottom, for those who need to plot F which are not equal to the
even spaced F bands.

To draw a response graph, the gear is all set up with pink noise levels sounding
loud enough to drown out most other noise. A level measuring about 85dB on a
cheap SPL meter should be OK.
The bandpass filter is set for the reference frequency of 1kHz, and output of the
mic amp is adjusted to give a meter reading of what is 0.0dB with meter needle
in the centre needle REFERENCE position.
Some slight change in metered level will be seen to occur because the level of the
filtered noise centered around 1kHz varies in amplitude at varying rates between
very low F and much higher F.
The peak detector circuit produces a Vdc which floats on the peaks in the noise
and the diode plus RC deector circuit cannot have a time constant that is too high
lest the meter take to long to settle at a value which best describes the average
peak levels. So one has to get used to a meter needle hovering around a centre
reading position.
But it is easy to adjust the reference level so that the needle appears to hover
around the 0.0 center meter position. The bandpass filter may then be switched up
or down along the audio band and at each selected F the meter needle will hover
around a center value which can be read off easily despite meter needle hovering
maybe +/- 1 dB. I take about 15 minutes to switch along all frequency bands and
place a dot on the graph with a pencil at a given dB level within +/- 0.5dB.

The use of the above method becomes easy to use with practice without measuring
any actual voltage levels in volts. One only has to set up the meter to read the
0.0dB meter level for 1kHz, then just plot where the needle goes along the band.

Response testing of each speaker in a stereo pair should be done with each in the
same room position relative to the microphone. In an ideal world, the two
speakers should have very close responses, regardless of whether the response
is very flat or not. Where large peaks and troughs are measured exceeding +/-3dB,
or that bass levels are poor in comparison to MF and HF bands, big one may find
that the crossover filters and the relative phase of drivers must be investigated
for shortcomings.

It is extremely rare to find that any mass marketed speaker will give an ideal
response. If there a response graph shown in a user manual or shown on the
rear panel of a speaker, I am usually led to believe how well the makers have
lied about their product. It is not uncommon for speakers to have a "loudness"
characteristic with boosted bass and treble so that people in shops will be well

The speaker impedance at all F should also be tested and graphed, but that
operation uses a completely different technique to be described elsewhere.

Faults with crossovers or driver units may be discovered by measuring the

The response measuring method may be used to improve a given pair of
speakers or while producing a new pair of speakers.

Testing each driver.
The response of each driver should be separately plotted during work to
build new speakers or reform old speakers.
This must be done without any crossover present and at low enough levels
to not damage tweeters with LF signals. Dome tweeters are very easily
damaged, yet it is important to plot their unfiltered response at least down
to 1kHz, when the intended crossover point is to be say 3kHz.
To avoid damage, a capacitor of 47uF can be used in series with a 4 ohm
tweeter which acts as a high pass filter with pole at 845Hz. For 8 ohms,
use 22uF. From the responses measured, one will begin to see what is
possible to achieve with crossover filters. The individual driver response
with driver mounted within the intended enclosures and when powered
by an amp with Rout < 1.0 ohms may give puzzling variations in response
levels. Bass, midrange and treble drivers may show quite large variations
of average levels.

The measured impedance for each driver in its intended enclosure must always
be measured carefully, but this is not done using pink noise signals and will be
discussed elsewhere.

Once the response graphs for all drivers is plotted without X-over filters, the
amount of attenuation measured in dB which is required for each driver to attain
a flat response may read from the acoustic response graphs. Usually the bass
will begin to roll off somewhere below 100Hz. The -1dB level of bass response
will set the levels of all frequencies above. Depending on how well the bass
driver suits its ported or closed box, one may be lucky to find the -1dB
bass response extends down to say 25Hz, and this may be 6dB lower than
midrange response at 1kHz. In fact the bass driver itself may produce a
fairly flat response from 30Hz to 100Hz, then show a rise in response level
of 6dB between 100Hz and 1kHz, and between 500Hz and 3kHz, bass
response may have a number of peaks and troughs to +/- 9dB, caused by
"cone breakup" where the cone deforms dynamically to produce very odd
behaviour. It is for this reason I don't like 2 way speakers with just a bass
plus midrange driver with a tweeter handling all F above 1kHz.
The LOW bass frequencies is usually where speaker sensitivity is at its
very worst, but for a flat response the low bass should set the reference
levels for all other frequencies. The crossover filters have to flatten the
acoustic response within the intended band for wach driver, and eliminate
some unwanted peaks, and make average levels for all drivers flat between
say 70Hz and 20kHz. The crossovers may also need to have resistance
attenuation networks to lower midrange and treble output levels, yet provide
enough resistance termination to damp resonant behaviour of LC filters.
Combined effects of crossover filters must not present impedance that is too
for an amplifier to safely drive. So there are suddenly very many things
to consider after making response tests.

As testing proceeds, the evidence about the real nature of the speakers is
accumulated to be taken into consideration when designing the Xovers.

Testing Speaker Impedance.
This requires a power amplifier, 1k x 5Watt resistance, CRO, ie,
oscilloscope, and good wide band millivolt meter. DMM are useless above
2Khz to measure signal voltages. Good voltmeters able to measure between
2Hz and 200kHz and all levels from 1mV to 1,000Vac, but are difficult to
find, and the CRO may be used to measure voltages well enough once it is
adjusted for a reference level at 1kHz. The signal generator must have
variable F and be fairly well calibrated for F between 10Hz and 50kHz at
least, and once adjusted for level, must give a flat response of sine wave
within +/- 0.25dB. Most good signal generators have Rout at 600 ohms,
and have several ranges between 2Hz and at least 200kHz. One might
easily make such an oscillator with a Wien Bridge circuit using a single
opamp and a good double gang log pot of 50k, lamp globe used for Vo
stability, and with THD < 0.5%, which is adequate for all speaker tests.

Measuring Speaker Impedance.

The above shows a set up for measuring speaker Z. The power amp is fed
by a signal gene to produce a flat response signal of 10Vrms at the output.
This is fed to a 1k0 series R to the speaker under test. A CRO may be used
to measure relative impedance levels. Actual impedance ohms may be
calculated well enough by measuring the voltage across speaker and x 100.
So if VLS = 0.08Vrms, then Z = 8 ohms. One can use a DMM to measure
all voltages below 1kHz, but most DMM become unaccurate above 1kHz.
The CRO trace height may be set for 1 graticle = 0.08Vrms, or at frequency
below 1kHz speaker has Z = 8 ohms, which is usually easy to find with an
8 ohm speaker which often has Z at 8 ohms somewhere between 200Hz
and 1 kHz. If one moves to another F and trace height = 2 graticles, then
voltage = 0.16Vrms and Z at the new F = 16 ohms.

Most speaker impedances measured will be between 2 ohms and 50 ohms.

Measuring Crossover Response.
So far we have dealt with measuring a complete speaker's acoustic response,
or acoustic response of an individual driver. We have also dealt with the
method of measuring the impedance of complete speaker or a driver.

But we also may need to measure the frequency response of the signal applied
to speaker drive units when the crossover filter is in place. If the speaker is one
we are to repair or upgrade, the schematic of the existing X-over filters must be
plotted carefully by reading all L, C and R values on the X-over filter board/s,
and then figure out all track connections. Its sometimes not easy when tracks
are hidden from view, and where those absoluely attrocious variable driver
level potentiometers or switches plus R have been used.
All speaker midrange and treble attenuators of any kind are best removed from
all crossovers, but we will still need to know what exists before moving in with
soldering iron and long nose pliers.

Most good 3 way hi-fi speakers will have a separate X-over filter for LF, MF
and HF drivers. Where attenuators have been used, set them all for their mid
position wherethe manufacturer was hopeful that the acoustic response
would be flat. The Z for the whole speaker can be measured and then for
each driver with X-over in place, then for each driver with no X-over.

The test set up above may be used for plotting speaker voltage responses
after the X-over.

The amplifier output must be adjusted right down to maybe 0.2Vrms.
The 1k0 series resistance is shunted with a strap. Loud Sound will be heard
from the speaker, and we might need ear plugs if annoying.
The CRO must be set up to read the voltage applied to each driver and for
this some access is needed to the X-over filters which are buried inside the
speaker enclosure. To gain access, usually the bass driver is removed, and a
coax lead with aligator clips to outer shield and inrrer wire is threaded through
low bass port hole and to wherever we wish to measure on the X-over board,
rather than to each driver whose terminals may be inaccessable while drivers
are screwed to the box. It is probably that wires to drivers from X-overs
may need some insulation removed to clip onto suitable bare metal. One way
or another, suitable connections must be found. Care must be used to ensure
the 0V line from CRO to connect to the 0V line of LS and amplifier.

Once access is made to speaker driver signals, the bass driver may be screwed
back in with just two screws. This is OK for testing bass units. Once bass units
have been tested, the bass unit may be left removed because mid and treble
will have their own small enclosures and be unaffected by the presence of
bass speakers and the bass X-over filter. This assumes though than the X-over
for each driver is an entirely stand alone type, and not a series type where
mid + treble unts are fed with signal across an inductor in series with bass,
and bass is fed with a signal across a shunt capacitor.

The Series Type of X-over is a veritable nightmare to ever try to optimise.
The only reason it is has ever been used is because fewer L,C&R parts are
required, and company accountants dictate such rubbish be used to save
costs of production. Usually when a speaker becomes a 3 way, nobody
ever bothers with series X-overs, and what is mostly used are individual
X-over filters which are wired in parallel to the amplifier.

Once the connection to a X-over speaker terminals has been made, the voltage
maybe monitored on the CRO. A dual trace CRO is handy, and the amplifier
signal of 0.2Vrms can be displayed on one trace so it takes up 1/2 the screen.

The other trace is also pre-set to the same amp voltage before connection
to a driver voltage. When this lead is connected to a driver, the attenuation
by X-over and series resistance by the driver can immediately be read by
comparing the two traces. Any strange distortion or peaks in response may be
easily plotted on a graph in 3 dB steps while the F at the sig gene is adjusted.

The sig gene with pot F adjustment is infinitely variable, and peaks in response
may be "tuned". Such peaks in response often correspond with a drop in the
impedance "looking into" the filter especially if is an LC second order type which
requires adequate terminating R at filter output lest the L and the C act as a
very low Z resonant shunt circuit at the Fo, resonant frequency which is also
usually the wanted crossover F pole. Peaks in signal response giving low Z are
an unwanted tradgedy, and are often the reason for "difficult to drive"
speakers which can kill amplifiers. If crossovers for 6 ohm bass and 6 ohm
midrange both have poles at say 250Hz using 2nd order LC filters, Z at Fo may
fall to 4 ohms or lower for both filters. Thus the total Z at 250Hz could become
2 ohms, and this is where most audio energy exists, and such an arrangement
will kill some amps. Therefore the X-over responses and total impedance
character must be carefully measured out and many sheets of notes and
graphs be made lest we waste our time to get good sound. I laugh at people
trying to build speakers by guessing LC&R values, and using only their ears
to monitor their results. Sometimes they get their nose involved - smoke
has that certain aroma indicating serious shortcomings in their methods.

From what I have said so far, there are many mass produced speakers
which may benefit from complete re-engineering of crossovers and the
method is no different to building a new set of speakers.
In the 2 years up to the end of 2011, I re-engineered several sets of
speakers including smaller sized 3 way Magnat speakers which had
fused 5 out of 6 drivers including both 9" bass drivers and both.
X-over filters had burnt and melted inductor coils where wire size
was far too small. I also re-engineered a pair of large 1975 AR9 4 way
floorstanders which had extremely insensitive 3.6 ohm 11" twin bass drivers,
but with all other drivers being much more sensitive and of uneven impedance.

In the case of the Magnat speakers, all drivers and crossover parts were put
out with the recyclables, and all that was used from the original speakers was
the rear, sides and top panels of the box. A new front baffle was cut from
33mm thick MDF and painted to match original colours. A 3 way design
using Peerless 200mm LF, 120mm MF and 25mm HF all made in Denmark
were used. I made up new crossovers with coild having thick wire and I
better capacitors througout. The crossovers were built using plywood
panels with 16mm x 4g brass plated wood screws used for terminals for
solid copper wire "tracks" or links between components. X-overs are
easily removed as they are held in by wood screws to the rear of the box
with a layer of foam to stop vibration. Such X-overs are vanished well
and all parts well mounted, yet easy to service. I don't use crummy tag
strips or printed circuit boards. The AR9 were also done the same way
after ditching the attrocious 1975 engineering.

To summarize the re-engineering, I can list the procedures :-

(1) Decide what parts if any are worth re-using.

(2) Usually the enclosure is in good enough condition to allow an effective
outcome. But in some cases the enclosure has begun to fall apart if made
pine based particle board with plastic false veneer covering. Such speakers
are condemned, and the owner advised he is wasting my time.

(3) If a total rebuild is required, A set of Peerless drivers available from
http://www.wescomponents.com.au is considered. The range made in
Denmark are remarkably good, and offer very good quality between the
poor quality cheap asian made drivers and the overly expensive hi-end
drivers made by Scanspeak or others. While the exotic hi-end drivers
are nicely made I doubt the huge cost is justified. But I will use whatever
high priced drivers anyone is willing to pay for if they ask for them. 
Nobody ever does does though. I believe the sonic outcome is mainly
determined by the man who designs and instals correctly rather than by
the exotic quality and high prices of divers.

(4) Once the selection of drivers has been made and the set received from
the supplier, the front baffle is prepared or at least altered for new drivers
to be installed. A typical driver set would be 200mm cone LF, 125mm cone
MF, and 25mm dome HF. A separate new sealed enclosure for the rear of
the MF is usually needed to allow the MF to work down to 200Hz.
rear enclosure usually may be from 2 litres to 5litres, may be a simple box
using plywood, or a PVC tube cramped between rear panel and front panel
all screwed and glued, and well stuffed with acoustically absorbant polyester
wool. The enclosure is assessed for a suitable match for a ported reflex
design. The 200mm Peerless give excellent bass response with enclosure
volume as low as 50 litres, or less than 2 Cu.ft. ( 28L per Cu.Ft. )

(5) Before deciding on a bass speaker driver, the box volume is measured,
and the probable outcome is explored using a box matching program WINISD.
I find the program does predict the bass response well, even with sub-woofer
designs using the rugged 300mm XLS Peerless sub woofer driver.

(6) Following the fitting of all drivers, the impedance of each driver is
graphed to include whatever resonant modes may be present, and to
detect what may be the best crossover frequencies. Usually I shall
choose bass cut off at 300Hz, midrange cut off at 200Hz and 3kHz,
and treble cut off at 3kHz. The bands of F chosen will easily be
by the chosen drivers. The mid band impedance for each driver is

(7) From the impedance curves drawn for each driver, the F where
Z has increased +3dB due to the series inductance of the voice coil is noted.
It is very desirable that the speaker driver act as a resistor at in its
upper band region thus ensuring predictable attenuation with L and C
filter elements. To neutralise the effect of voice coil inductance, a Zobel
R&C network should be connected across each driver. Such a network helps
to damp the oscillatory speaker behaviour.
Suppose a bass speaker has a flat region of Z between 150hz and 300 Hz,
and is 6.5 ohms. Suppose the Z rises to 1.4 times the 6.5 ohm value at
600Hz. To calculate the Zobel R&C values needed, R = the nearest standard
R value above the mid band Z, in this case it will be 6.8 ohms. This resistance
will at least be a 10 Watt wire wound type.
The C value in uF = 159,000 / ( 1.41 x Z x Fo ) where 159,000 and 1.41 are
constants for all equations, Z is the midband speaker Z, and Fo is the F where
Z has risen to 1.4 x lower midband Z, In this case we get
C = 159,000 / ( 1.4 x 6.5 x 600 ) = 29uF. This is an awkward non standard
value, but a 30uF polypropylene motor start cap rated for 250V would be
fine. Alternatively, 2 x 10uF NP electrolytics could be used with an 8 uF
or 10uF polypropylene all in parallel.
When connected across the bass speaker, the impedance should remain at
6.5 ohms for all F above the 6.5 ohm mid band value at 200Hz.

The procedured of placingf Zobel networks across each driver is known as
"impedance equalising". It is often ommited where 2nd oder or 3rd order filters
are used, but I always use such Zobels.

(8) Following  connection of Zobels, the unfiltered response of each driver
is measured using a pink noise test from a power amp. The responses are drawn
on the same page as each driver is tested separately, so that it is fairly easy
to see what attenuation is required for the most sensitive speakers, and if the
band response may be optimally flattened, without lowering Z excessively.

(9) Crossover values are chosen. I have found the chart below indicates
what is required.

To Loudspeakers 1, New,

To Loudspeakers 2, DIY,

To Loudspeakers 3, crossover filters.

To Loudspeakers directory,

Back to Index Page.