Fan noise and PWM

silvervarg · Post by **silvervarg** » Wed Jan 07, 2004 7:36 am

I could not find any usefull information about fan noise for fans when driven with PWM. I know lots of people out there use Speedfan or similar to control fans and many hardware fan controllers use PWM as well.

So please post your findings here.
If possible supply this information:
Fan used. (Size, Brand, model number etc). E.g. 80mm Panaflo L1A
PWM duty cycle. E.g. 40%
PWM frequency if known. E.g. 35 Hz.
Noise other than airflow noise. E.g. Some noise, similar to L1A@6V
RPM if known. E.g. 948 RPM

dukla2000 · Post by **dukla2000** » Wed Jan 07, 2004 12:03 pm

I gave up on PWM when I discovered some of my fans made horrid noises at low duty cycles. I was playing with Speedfan on an ASUS A7V333 mobo that uses an ASUS proprietary PWM controller (that mostly emulates a Winbond xyz). Anyway, it gave 16 steps of duty cycles from 0 to 100%, and I had the impression it was 50Hz.

However at least with my 80mm Papst (NGL & NG) and 120mm NMB and Evercool I got really bad motor growls when dropping the duty cycle down below 50%. At which stage I read some manufacturer doc (Sunon or NMB) that said standard DC fan motors are not designed to handle PWM and could be noisy (or something similar) which is why they (& Papst) produce PWM specific versions.

So now I just undervolt things: a shame not to have software control but... (I know in another post here Dorothy/jafb2000 said this motor noise problem is a factor of the PWM frequency: if I had a more sophisticated controller & s/w I could probably tune things for each fan but ...)

Cob · Post by **Cob** » Sun Jan 11, 2004 11:20 am

I've experienced the same thing with my fans.
I have a Noise Isolator PWM fan controller, and have 2 vantec stealth 120mm hooked up to the first channel, 2 panaflo low 80mm hooked up to the second channel and I have my Zalman 92mm hooked up to the third channel.

I've found that even the awesome Panaflo Low fans get a nasty motor whirr when at the low speed, but then if I just undervolt it, then the whirr goes away.

However, I really like having the control over the fan's speed.

So, my question: Is there a fan controller out there that has the ability to handle a fair amount of fans, and has variable voltage control, that DOESN'T create the motor whirr?

silvervarg · Post by **silvervarg** » Sun Jan 11, 2004 11:58 am

I am trying out the fancontroller that was made by Fancontrol. I you have loads of time you can read his thread, now 10 pages long...
I will get more fans to try out with that controller, most will be 80mm fans.

So far it seems that high quality fans do fine with that PWM controller and low quality fans have lots of bearing noise with this controller also has lots of bearing noise when running with voltage regulation through a fanmate.
I have tested with the 24V M1A.
I have not tested with the L1A yet, but I will hopefully do that within a week. I will also test with 92mmL1BX.
I hope to pick up an 80mm ADDA and at least one 80mm Papst soon.

Some other test seem to indicate that PWM has slightly more problems as fan size grows. I have no plans on testing this out since I would need to get lots of fans I have no use for to test it.

Post by **Ralf Hutter** » Sun Jan 11, 2004 12:04 pm

Cob wrote:So, my question: Is there a fan controller out there that has the ability to handle a fair amount of fans, and has variable voltage control, that DOESN'T create the motor whirr?

Sure, this one, this one and this one.

Putz · Post by **Putz** » Sun Jan 11, 2004 12:05 pm

My own experience (obviously not the rule, but a sample nonetheless) has shown that, on my sister's laptop, the 52mm CPU fan responds perfectly to PWM, while the dual tiny fans (25mm?) squeal and tick like crazy at any values except 0% and 100%. Not sure what the frequency is.

Cob · Post by **Cob** » Sun Jan 11, 2004 1:03 pm

Thanks for the speedy - and exceptionally helpful - reply there Ralf!

silvervarg · Post by **silvervarg** » Thu Jan 15, 2004 4:52 am

Great to get some input on a few more fans Putz.
It would however be a lot more valuable if you could get some info on what fans that was. I hope you could at least get the brand and RPM at full speed.
(I don't think the 25mm fans are very interesting.)

silvervarg · Post by **silvervarg** » Sun Jan 18, 2004 7:27 am

I did some more testing on the Zalman 80mm fan. It does work remarkably good (compared to its price) when regulated with voltage (through a Zalman fanmate).
It is also remarkable how bad this fan reacts to PWM driving. On all the PWM setting I tried this fan caused lots of vibrations.

Inexplicable · Post by **Inexplicable** » Sun Jan 18, 2004 8:39 am

According to the datasheet, the Winbond w83627 I/O chip found on many motherboards has a default pwm frequency of 23.43 kHz. It has a number of different settings, all in the kHz range. I'm driving my cpu and nb fans with it and haven't noticed any noise issues whatsoever on any duty cycle setting.

I think the trouble with fancontrol's controller circuit is that it uses the AMD1027 chip, which has a ridiculously low pwm frequency, less than 100Hz on all settings according to the datasheet.

In general, I would guess that small fans are more susceptible to pwm noise than larger fans which have more inertia. With a high enough pwm frequency this shouldn't be an issue.

silvervarg · Post by **silvervarg** » Sun Jan 18, 2004 12:27 pm

That sound awfully high frequency for PWM.
According to Cpemmas site :
At low frequencies the motor speed tends to be jerky, at high frequencies the motor's inductance becomes significant and power is lost. Frequencies of 30-200Hz are commonly used.

So the 20-25 kHz range sounds way off to me. Also with that high frequency you would loose the torque advantage with PWM.
Do you have a link to that datasheet?

What fans do you run on Winbond chip that works fine?
Do you know the lowest PWM duty cycle they will start on? Preferably with the RPM of the fan on that PWM duty cycle.

In general, I would guess that small fans are more susceptible to pwm noise than larger fans which have more inertia. With a high enough pwm frequency this shouldn't be an issue.

This was my initial guess as well, but it seems that reality point in the opposite direction. I have no idea why, so any theory would be appreciated.
I still gather more facts, since I have too little to get reliable results yet.

Inexplicable · Post by **Inexplicable** » Sun Jan 18, 2004 1:36 pm

Ok, I'm certainly no expert on this. I could be misreading the datasheet. Look for PWMOUT1/2 Clock Select. Admittedly the table is a little unclear as it lists each frequency as "nnnK Hz" rather than "nnn kHz". I took the K to mean thousand but it could just be an editing mistake. The fans I have are nothing uncommon, a 92 mm Zalman on the 7000cu and on the NB the stock 40mm fan that came with the Abit board.

If a typical pwm frequency truly is < 200 Hz, it's clearly quite close to the rotational frequency of a typical fan. I can imagine how that could easily produce vibrations synchronized with the rotation if the force applied on the axle is not quite orthogonal in all axle positions.

silvervarg · Post by **silvervarg** » Mon Jan 19, 2004 4:13 am

Thanks for the link. I do agree that it does look like the PWM modulation frequency of the Winbond chip is 2.93kHz to 46.87kHz. Perhaps someone with good equipment can measure and see what the frequency really is.
It seems like the Winbond chip supports 2 PWM outputs, but there is some kind of support for a third PWM output, but I can't see any usefull info on that third PWM. E.g. no place where you can set its PWM duty % etc.

Jan Kivar · Post by **Jan Kivar** » Mon Jan 19, 2004 5:35 am

I think that the datasheet has correct values.

Looking at the datasheet on p. 33/37 (doc/Acrobat) You'll see an example circuit, which utilizes a cap in parallel with the fan. This is to smooth out the high PWM cycling. But, using the cap requires a fan which pulls (sinks) enough current (to drain the cap). Using a small-amp fan (like Panaflo/Papst we all like), the cap just doesn't discharge enough to enable enough voltage swing to the fan.

Here's a post by cpemma about this, followed by my post.

I think this could be circumvented by using a "right" size cap for a specific fan.

Cheers,

Jan

Inexplicable · Post by **Inexplicable** » Mon Jan 19, 2004 9:55 am

I suppose this is only a problem if the cap on the mobo is too large for the fan. Too small a cap would simply provide poor smoothing but would not otherwise interfere with the pwm. The latter problem could be compensated by wiring an additional cap in parallel with the fan.

silvervarg · Post by **silvervarg** » Tue Jan 20, 2004 2:23 am

A small cap to smooth out PWM is a common trick to lower the effects of clicks. Going with a higher frequency is another trick to do essentially the same thing. Using both should give a rather smooth curve, but it does have some disadvantages.
It essens you get a voltage control circuit with some ripple on the signal and probably lower torque. Also you need to have a quick FET or similar to keep down the generated heat or use a heatsink on the FET. This will only have a large effect if you have a high power fan, so another way is to just specify a low limit on how much power the fan may use.
Perhaps this is still cheaper than going for a voltage regulated fan controller, but IMHO you have lost all the benefits of PWM except perhaps the price.

Is the winbond chip used on most motherboards?

Inexplicable · Post by **Inexplicable** » Tue Jan 20, 2004 12:00 pm

Winbond chips are probably the most common by a fair margin but there are a number of other chips as well (e.g. National Semiconductor, VIA, ADM, ...). The MBM site lists the chips used by a large number of different motherboards (not all of them support pwm).

I think I've figured out why the Winbond frequencies seem high. Apparently they are talking about the frequency at which the pwm circuit itself is clocked. Since there are 256 different pwm levels, the pwm output frequency is actually input frequency / 256. That gives an output range of 11..183 Hz. Make sense?

silvervarg · Post by **silvervarg** » Tue Jan 20, 2004 1:34 pm

Ah, now it makes sense. Thanks for clearing that up. Did you find anything about that in the datasheet?

Inexplicable · Post by **Inexplicable** » Tue Jan 20, 2004 1:58 pm

Not exactly but their W83L785R datasheet is a bit more explicit.

bigred · Post by **bigred** » Tue Jan 20, 2004 2:10 pm

I have the original Vantec Nexus. It uses PWM, yet I do not find that it causes the fans to make any additional noise. However I do find that the capacitors buzz when you have a fan turned right down. When I have all the fans turned right down to ~5v the buzzing is quite loud and can be hard from up to 2 meters away, albeit quiet. From seating position its very distracting.
I also notice this noise on 2 other's, so I know its not just mine.

I'm not sure it this is the sound people were referring to, but I am quite certain that its not coming from the fans.

silvervarg · Post by **silvervarg** » Thu Jan 22, 2004 4:24 am

I did some experiments with my new panaflo's that I got from Dorothy.
The results are far from what I was hoping for.
I did get a few 24V M1A, some normal L1A and some 92mm L1BX.

The L1A does click and vibrate when running on PWM, but not when voltage regulated.

The 24V M1A clicks a little and very little vibrations when running on PWM. Clicking and vibrations are a lot less when voltage regulated.
The clicking and vibrations depends on the PWM frequency. It seems that slower speeds like lower PWM modulation frequency than higher speeds for optimal performance.
Frequency of 11Hz to 44.1Hz works best. I only got more noise when I turn the frequency up to 58.8 or 88.2Hz.

Now to the stunning thing:
I did some tests with my last shipment of Panaflo's a few months ago. I only had 24V M1A's in that shipment. The "old" 24V M1A works a lot better on PWM than the new ones. Both old and new fans works great on voltage regulation.
The difference is that the fans that work fine on PWM are made in Japan, while the fans that don't like PWM are made in China!
This could either be due to production quality or possibly that the Chinese fans are fundamentally different for some reason.

If anyone living close by (Sweden, near Linköping) have a Panaflo 12V 80mm L1A that is made in Japan I would like to swap for one of my new L1A from China so I could test this theory.

I still have to do some more test on my 92mm L1BX, but from initial testing I think the results will be the same as with the other fans.

cpemma · Post by **cpemma** » Thu Jan 22, 2004 4:21 pm

Inexplicable wrote:I think I've figured out why the Winbond frequencies seem high. Apparently they are talking about the frequency at which the pwm circuit itself is clocked. Since there are 256 different pwm levels, the pwm output frequency is actually input frequency / 256. That gives an output range of 11..183 Hz. Make sense?

That explains some things that were puzzling me. As you say, the fast clock is more like a timebase. So for a 70% duty cycle, the output goes high & switches on, counts up to (256*0.70) = 179 fast cycles, then output goes low & switches off for (256-179) = 77 fast cycles.

Much better explained in your link than the pdf I had.

On "ideal" frequency, Micrel and Microchip both suggest 30Hz for PC Brushless Fan Control using their ics, and they've got the research facilities

Bat · Post by **Bat** » Thu Jan 22, 2004 4:26 pm

I still have the suspicion that the best way would be to have the PWM frequency the same as the fan switching frequency though (rotational frequency x number of poles), with the pulses timed to arrive at the best phase of the fan switching cycle.
I don't know how you'd achieve that, but it could be interesting.

Jan Kivar · Post by **Jan Kivar** » Fri Jan 23, 2004 1:33 am

cpemma wrote:
Inexplicable wrote:I think I've figured out why the Winbond frequencies seem high. Apparently they are talking about the frequency at which the pwm circuit itself is clocked. Since there are 256 different pwm levels, the pwm output frequency is actually input frequency / 256. That gives an output range of 11..183 Hz. Make sense?
That explains some things that were puzzling me. As you say, the fast clock is more like a timebase. So for a 70% duty cycle, the output goes high & switches on, counts up to (256*0.70) = 179 fast cycles, then output goes low & switches off for (256-179) = 77 fast cycles.

Much better explained in your link than the pdf I had.

On "ideal" frequency, Micrel and Microchip both suggest 30Hz for PC Brushless Fan Control using their ics, and they've got the research facilities

Looking at the pages 31-32 in the doc Inexplicable linked, You'll see IMO the "right" equation to calculate the PWM output frequency. On page 31 is said that there are two PWM input clocks; 125 kHz and 1 MHz. On page 32 is the prescaler setting, along with the equation.

So, the min output freq. for 125 kHz input clock should be 125 kHz/127/256 ~3.84 Hz, and the max. 125 kHz/1/256 ~488 Hz, as calculated in the doc.

For the input clock at 1 MHz, just multiply the values by 8, giving ~31 Hz; 3900 Hz, respectively.

Now, if I screwed up the calculations/theory, please let me know. I'm trying to design/build a fan controller, so I should get these right if I'd ever want it to work....

Cheers,

Jan

Jan Kivar · Post by **Jan Kivar** » Fri Jan 23, 2004 1:39 am

Hmm... a thought just crossed my mind: Were You taking the values from the other datasheet?

Jan

Inexplicable · Post by **Inexplicable** » Fri Jan 23, 2004 9:33 am

Yes, I was.

I think it's plausible that the 256 divisor also applies to the w83627f, even though the datasheet doesn't spell it out. That one doesn't have a prescaler.

cpemma · Post by **cpemma** » Fri Jan 23, 2004 2:12 pm

Jan Kivar wrote:...So, the min output freq. for 125 kHz input clock should be 125 kHz/127/256 ~3.84 Hz, and the max. 125 kHz/1/256 ~488 Hz, as calculated in the doc.

For the input clock at 1 MHz, just multiply the values by 8, giving ~31 Hz; 3900 Hz, respectively.

Just varying the clock frequency wouldn't vary fan speed AFAICS. I read the 1-256 count as being the way of setting a duty cycle. So the fan PWM frequency would be a fixed 488Hz if the 125kHz option were chosen. I'll read it again...

silvervarg · Post by **silvervarg** » Sat Jan 31, 2004 3:55 am

I did some experimentation with capacitors on the fan wires (from +12V wire to GND wire) when doing PWM.
If you got a way to large capacitor it will keep enough charge to power the fan during most of the PWM-off-cycle, so you get close to full 12V load on the fan.
A small capacitor does however just smooth out the PWM cycle, and the clicks magically disappear without affecting fan speed much. Note that I am just taking about the clicks that appeared from PWM controll, and not "normal" bearing clicks.

From my tests it seems that a 2.2uF capacitor works great.
Even an 1.0uF capacitor take away most of the PWM clicks.
Bigger caps will just ruin the PWM more, and not add any benefits over the 2.2uF.
I tried with a 92mm L1BX Panaflo 0.15A and 80mm L1A.
The 80mm 24V M1A from Japan does not seem to have much PWM clicks at all, so adding a cap does not change noise much.

jafb2000 · Post by **jafb2000** » Fri May 07, 2004 6:10 am

PWM problems occur on several levels:

1) PWM Frequency choice
o 2kHz -- relatively common in industry, but an audible
o 20kHz -- increasingly common trying to get above acoustic range

2) PWM interaction with fan microprocessor
o Top fans don't use a simple transistor switching system
o They use a microprocessor
---- fan-start & restart, timer
---- soft-switching - like some hard drives have soft seeking re accel/decel
o As the PWM constantly chops up the voltage it causes havoc
---- Vs rise - fan re-enters timer-delay-to-fan-restart mode
---- chopping - interrupts the soft switching by the microprocessor
o Net result is you can get lots of weird & wonderful noises

3) PWM interaction with fan bearings
o Free play decreases Papst -> NMB -> Panaflo
o As bearings load & unload a click can be heard like on fan starting
o The click will occur at some frequency below PWM
---- mainly where the motor loses torque re least magnetic interaction
---- and where the PWM frequency is overlaid on that of soft-switching &
fan-restart circuit triggering in particular (which has a cut-out timer)
o So predicting how a fan will perform by bearings is futile
---- just that it's another variable

Overall, it comes down to "suck n see".
o Varying PWM frequency does help
---- anything below 1-2kHz is generally disliked, 50Hz is mains hum
o Varying PWM duty cycle only between 50-100%
---- below 50% and the fan restart circuitry will have kittens
o Trying smoothing capacitors
---- not a bad idea, remember the 12V supply on a PC is noisy anyway

Papst's varioflow fans (eg, MV code) only do 50-100% for this reason.

Frankly, a variable voltage regulator is not a bad idea - dissipation is
not that much for typical low-noise PC fans re 35ma at 6V, 70ma at 12V.

Fan bearings may also not tick on rotation, but on thrust.
When you slap a Panaflo hard, you'll hear a loud clack - that is contact
being made between the hydrowave fluid-bearing capsule and a (very!)
hard steel floating shim and a brass thrust bushing. That is there to give
the fan a rugged rating far beyond even NMB ball bearing, since the aim
of the bearing was ruggedness from Broadcast Video Recorders, to in
some applications hard drives, to real Industrial & Medical arenas too.

Quite hard to get PWM to involve significant axial thrust, but it is possible.
Cheapy fans will just double-spring the main fan bearing, which is good
for noise but the play/axial-runout when running toasts the bearing either
in high/low temp or with age eventually. Exactly what the Papst Sintec,
Ball, NMB Ball & Panaflo ball/pivot/hydrowave bearings aim to avoid.
The hydrowave is the highest precision bearing of all, but well proven
with field data of 1990-1996 on the earliest bearing having 0 failures
out of 20M units, and oil usage tests indicating life >114,000hrs at 70oC.

PWM will get more problematic with larger diameter fans, and especially
the Panaflo since the hydrowave bearing (& pivot altho you would do well
to find one in retail or OEM as it's the next fluid bearing down from them)
is so large as a result of the required aspect ratio. Hence Panaflo will
never do a 120x25mm fan with it - and their big 160+mm fans use ball.
The 120mm M1A with it's huge heavy magnet tends to growl with PWM.

PWM comes down to testing the proposed fan individually.

Temperature based PWM or voltage regulation is somewhat poor, the
ideal which the industry is moving to is current based modes of control:
o Temp thermistor control is a lagging, high latency & hystersis problem
---- CPU/PSU peak thermal loads can be instant
---- the fan conversely may take some time to react - if at all
o Current based management is direct instantaneous response
---- ie, the VRM circuit uses a shunt resistor & comparator to vary Fan Vs
---- some thing it may be adopted nearer 2006

It would be an improvement - not just for desktops, but also laptops.
Laptop fans have recently adopted a stepped speed control, but can have
a high temperature-led latency vs instantaneous & infinite current-led.

Current-management may be done by fans, or by chips on motherboards.
Will be interesting - as the value-adding fight plays out between the two:
o Top fan makers are eventually planning programmable fans
o Semi makers are planning better fan controllers

When solutions are outside the fan, they face the impossibility of coping
with all the various fan drive systems & bearing systems out there. Hence
PWM is mainly aimed at running fans at 80% to boost life, at least in terms
of reducing redundancy needed re lesser servicing cost but the cost of a
servicing person actually knocking out unrelated systems during a service.
For datacentres, over 35% of downtime is by someone servicing one rack
server only to knock out another or a rack power/communication system.

Panaflo & NMB had a Joint Development Programme running for some
time, but that still leaves two S.G. & A systems re two separate Co's. A
lot of duplicated operational work, like an electricity & gas utility Co's.
So as expected they merged in April this year - doesn't mean much yet,
eventually the name of both will change, probably to "M3" (unofficially),
and there will be more availability of some interesting fan-options/sizes.

Presently both companies are merging their processes and should be
complete by mid-summer - with far beyond five-9's on quality the result.
That is beyond 1 in 100,000 in terms of quality variation. I suspect a lot
of work is going on re production engineering of the hydrowave, which
has one of the highest precision requirements in the world - an order of
magnitude above even NMB ball bearings. So knowledge transfer both
ways with the merger to NMB from Panaflo & from NMB to Panaflo. As
Papst, Panaflo & NMB fight for the lowest sub-50ppm failure, it should
prove interesting. It should also boost availability of both brands.

A 60mm FBZ is on the horizon, but the main improvements are in the
actual production processes - NMB are very vertically integrated in that
they manage raw-material to product in-house, Panaflo need the plant
capacity of NMB despite brand new China plants & that at Osaka Japan.
Demand is outstripping capacity hence some model plain unavailability,
and particularly inability to order option codes with reasonable timelines.

The Panaflo 40mm will remain a ball-bearing, but that they hope to get
down in price - incredibly smooth & quiet, but very very expensive. The
key difference between ball & hydrowave is in dB(A) slip, ball will tend
to increase in dB(A) from day-1 over the service life whilst hydrowave
does not increase in dB(A) and particularly in actual acoustic spectrum.

Unfortunately, no 120x25mm which is a pity.
RS Components in the UK want about £26 for the NMB 25dB(A) 120mm,
so that hopefully will change - it's a great fan, but that's a bit OTT really.

Try NMB with PWM, they may work where the others don't.

Note that even Papst's own controllers, the PCM001 & System 3000 are
often listed with certain Papst fans but quite unusable with them re growl!
I suspect they may have been suitable in the past, but the continual fine
refinement of soft motor start & motor-ic's over times reduces that.

Fan's have a spectrum of pure -> white-noise, Verax have noisy bearings
whilst the blade design is slightly superior to the Panaflo FBL/FBA/FBZ.
PWM can play around with that white-noise spectrum very considerably,
and turn a pure overall perceived frequency of NMB/Panaflo into broad.

If you must have a PWM 120mm fan, the 4312MV Papst is very good.
It is 21dB(A) at 40cfm at ~25oC, and a real 21dB(A) not a fiddled one.
There can be some faint occasional ticking, but there is ticking & there
is ticking with it being mostly inaudible at a foot or so - and especially
if you have more than 1 fan in an enclosure (white-noise masks it out).

PWM is a technical solutions whose physical realisation isn't so predictable.

Skylined · Post by **Skylined** » Fri May 07, 2004 12:48 pm

I like voltage regulators.

http://www.gideontech.com/guides/var_speed/