Quieting Noisy Fans -- A HP 16702B Logic Analyzer

[TimInCanada]

I picked up one of these logic analyzers, and as Matt's Tech Page (http://tech.mattmillman.com/16702b/) describes it so aptly:

"It’s just a shame the rest of it makes a noise level somewhere between a hovercraft and a jet engine."

After turning it on, I concur.

The standard solution is to look for quieter fans with similar airflow rating, but after looking closely at this unit, it just isn't designed for good cooling.  The five noisy fans look like a quick fix to get the design out the door. 

Here's the general layout of the chassis with the cover off, looking from the left rear corner.  There is one fan on the left side.  The other piece with the two fans is called the center strut, and it goes down the middle of the chassis, on the left side of the card cage, blowing air over the plug-in cards.



Here's the motherboard at the top rear of the chassis.  The case side fan is on the left and the big black heatsink on the right is the CPU.  A small heatsink behind it is the video chip.  Only part of the fan peeks up over the circuit boards.



Oh, but it gets worse.  Here's the board-level view of the CPU heatsink. 



That daughter board has connectors underneath that completely block the area between it and the motherboard.  That ribbon cable on the left blocks airflow moving above the daughter board.  Needless to say, the CPU wasn't being cooled very effectively, even with a big, noisy fan.  I think we can do better.

I opted for PWM fan speed controller.  This one was $8 on eBay:



It controls two zones, reading the temperatures with two thermistors and providing a PWM signal to control fan speed in each zone.  There is some minimum fan operating speed, maybe 30% of full RPM.  You can set a low temperature where the fan speed starts to increase, and a high temperature where it reaches 100% RPM, apparently with a linear ramp in between.  It uses 4-wire, PWM fans.

There are cheaper fan speed controllers, but this one provides a digital readout of temperature and has an alarm if either fan stalls.

So for Step 1, we'll see what we can do about CPU cooling.  To exercise the CPU I did a "reignite", which is a reinstallation of the system software from the CD-ROM.  This takes about half an hour.

For the base case, a thermistor probe was placed between the fins of the CPU heatsink.  The machine averaged about a 285 watt power draw.  Temperature steady state was about 43.2C.  There were no add-in cards in it.

Next, a 80 mm 4-wire fan from a PC was put on the CPU heatsink, and connected to the controller.  A piece of a plastic DIP tube was cut to go over the thermistor in the heatsink to shield it from a direct blast of air from the fan.  The fan on the side of the chassis was disconnected.  On reignite the steady state temperature was about 28.3C and the fan speed was 1450 RPM.  Running just this fan at this speed showed it was barely audible, a huge improvement over the existing fan.

The thermistor was then epoxied to the heatsink:



The epoxy should allow heat to more easily flow into the thermistor, so the reading should be closer to the actual heatsink temperature.  Because the clearance between the top of the fan and the cover was only about 5-6 mm, I trimmed the top flange off the fan.  Hopefully this will let it breath better.  I also removed the heatsink and replaced the old, dried up thermal compound.



Another reignite, and the temperature was 36.2C with the fan turning 1570 RPM.  The heatsink didn't feel any hotter this time around, and I believe the higher temperature reading was due to the epoxied thermistor more accurately reading the actual heatsink temperature.

Also, feeling around all the boards after the reignite test showed that the video chip with the small heatsink was a little warm, but not uncomfortable to hold onto, and one chip on the SCSI board was not quite that warm.  Otherwise, all the other parts were at about ambient for the inside of the case.

Now we will keep in mind that I bought this unit "as-is" and non-functional.  The video display during the reignite was only text and it is possible that displaying graphics will cause the video chip to run hotter.  That's something to check when the unit if fully functional.

So, one noisy fan eliminated with better cooling of the CPU.  Not bad progress so far.

[TimInCanada]

On to Step 2: the two fans in that center strut.

These are the fans that blow air into the card cage.  This design actually turns out to have been a retrofit after the 16700 units were in production.  The original chassis had two fans on the side and none on the center strut.  There was a retrofit kit available and the 167xx card manuals caution to use them only in mainframes with the center strut fans.  The 167xx cards must have been found to run hotter than the 165xx cards.

Here is that center strut with the fans removed.  The left side goes to the rear of the chassis.



The rear fan also had two plastic card edge guides in front of it, not shown here.

I guess they designed that grille punched into the strut for the rear fan thinking that customers would pull all the cards and filler panels, turn the unit on, then reach in the card cage and stick their fingers into the rear fan.  But not the front fan.  Sometimes you wonder...

Anyway, on the cards I have most of the heatsink area is towards the rear, yet the rear fan has the greatest obstruction to airflow.  Not only is there that grille, plus two unused card guides on the fan intake side, but notice the largest diameter of the grille opening is less than the diameter of the hole for the front fan.  Some work with a hand nibbler eliminated the grille, and opened up the hole diameter to match the fan diameter.

The next problem is that big rectangular opening at the front of the center strut.  This matches one on the other side of the card cage, and is obviously left over from when there were no fans on the center strut.  The problem is that there is a low pressure area behind the fans, and a high pressure area in front.  The air will rather short-circuit in a path to the front of the cards, through this hole and back around into the fans.  I.e., further diverting airflow away from the heatsinks on the back of the cards.

It would be like trying to air condition your house with the windows open.  Sure, it can be done, but the air conditioner will need to be much bigger than necessary for the desired cooling.  So, let's close that hole:



A handy piece of aluminum almost covered the hole, so in keeping with the econo theme, kapton tape was used to close the remaining gaps.  The two white plastic card guides were not put back in. 

When the unit is assembled there is also a gap at the top of the strut which would allow air to short-circuit from the top card back to behind the fans.  I found some weatherstripping foam that filled most of this gap, but didn't touch the bottom of the motherboard.

Oh, one more thing.  See in the first picture that the open area from the fans to between the card guides is different for the five card slots?  This suggests putting cards that run hotter in slots 3 and 4 to get the most cooling air, then slots 2 and 5, and finally slot 1 which will get the least airflow.

The next question is, how much fan is now necessary?  The answer is, I don't know.  Probably much less than there is now.  I'm also going to take advantage of the fact that the was rated to run at up to 50C ambient.  I will never run it at such a high ambient temperature.  In my basement workshop it probably never gets above 25C.  That provides extra margin to work in.

You can slow down these fans by dropping their supply voltage.  As a first attempt to the right speed, I ran them on a bench supply and found their noise level tolerable at 6V, and they still put out significant airflow.  Putting five 10 ohm 1/4W resistors (which were on-hand) in series for each fan provided the right voltage drop without heating the resistors very much.  These are stuck to each fan with hot-melt glue in the photo above.

I let it idle with five cards in it for an hour or so.  (It's not working fully yet, but is drawing about 680 watts compared to its specification of 630 watts maximum, so heat production must be at or near maximum even while not taking measurements.)  Using a thermocouple on the outlet side at about the middle of the rear half of each card got these air temperatures:

Card 1: 17712A, 35.1C
Card 2: 17534A, 41.5C
Card 3: 17555A, 28.7C
Card 4: 17555A, 29.2C
Card 4: 17555A, 27.0C

Qualitatively, the air coming out of the front half of the card cage feels only a little above ambient while the air coming out of the rear half of the card cage is noticeably warm.  I'm going to conclude the airflow in the rear half of the card cage is too low.  So the next step will be to speed up that fan.

More to come,

Tim

[duak]

I've also converted some of my instruments over to quieter fans and it makes them easier to tolerate.

I have a prototype board from the early 90's that uses some CY100E484 ECL fast static RAMs.  If memory serves, these specified a lineal cooling air flow speed of so many feet per second to not melt down.  The board was designed with a temp sensor on one of the chips and during bench testing had a fan and shroud to cool it.  I wonder if the 16702B might not also need extreme cooling for some chips?

Cheers,

[TimInCanada]

I've also converted some of my instruments over to quieter fans and it makes them easier to tolerate.
[...] I wonder if the 16702B might not also need extreme cooling for some chips?

Hi duak,

Yes, it's certainly possible some parts may be specified to have particular airflow levels over them.  I'm doing something of an empirical approach, with the criteria being are any parts very warm to the touch.  If there's not enough airflow over a particular part, it should show it by its temperature.

With that last test mentioned, I shut the unit down after an hour or so then pulled the cards and felt around them.  Nothing was particularly warm.  Even the card with air temperature coming off at 41C didn't have anything too warm to touch comfortably.

[TimInCanada]

And on to Step 3: the power supply fans.

The power supply is behind the front panel.  It's rated at 700 watts and is surprisingly compact.



So compact that looking through with the fans removed, it's hard to see any daylight:





This is what the fans blow into.  The power supply is in two stages, the first making approx. 360VDC, then those six modules converting that to +5, -5.2, +/-12 and +3.3.  Those three brown filter caps are 220uF/450V.  Unfortunately, there aren't smaller caps with this rating that could be substituted to give less restriction to the airflow.  At the power inlet on the right it looks like some of the airflow is supposed to go through one set of perforations of that cage and out the rear perforations.

The fans are 60 mm Nidec TA225DC M33455.  There is a data sheet available at that gives the fan curve for this model:




The static pressure is the pressure rise across the fan.  The more restriction is placed on the airflow coming out of the fan, the higher the static pressure gets but at the expense of total airflow.  That inflection point is where the blades of the fan aerodynamically stall and cause an efficiency drop.  (And probably noise increase.)  It's best to operate to the right of that point.  The units are inches of water column, that is, how high a column of water will be lifted by this pressure.  For this fan, best running is with a static pressure of less than about 0.08 inches or 2 mm.  So not much.

The challenge is, there is no easy way to determine the flow resistance the fan will face.  If the flow resistance is lower than the fan can work with, the volume of airflow increases, which is a good thing.  More volume of air also means greater velocity which improves the convective heat transfer coefficient, which is also a good thing.  So reducing restriction to the flow of air is a priority.

One option is to replace the current 60 mm fans with quieter models.  Noctua fans get good reviews and their 60 mm model is rated at 29.2 m³/h at 2.18 mm H?O static pressure.  (Often manufacturers only give flow at one static pressure, rather than providing the full fan curve, making comparisons between fans more difficult.)  That translates to about 17.2 cubic feet/minute at 0.088 inches of water.  That's about 50% more airflow than the original fans at this static pressure.  And with a noise rating 10 dBA lower.  Running this model fan with thermostatic control would also let it run more slowly, producing less noise.

In addition, if the restriction of the airflow path is reduced, the fans can be turned down even more and give the same airflow.  Here's the inlet flow path to the fans:



Not very promising.  I'm looking at this unit as a long-term replacement on my bench for both logic analyzer and scope, so I'd like to go for as little noise as I can.  What about using one bigger fan turning more slowly?  Here's the 120 mm fan from the chassis side hung in front of the PS.



It will take a little chassis surgery, but could work.  A 120 mm fan will flow more air than two 60 mm fans at similar conditions.  In the Noctua fans, the F12 (highest static pressure model) flows about triple the air as the A6 into about 20% higher static pressure at the cost of 3 more dB.  It has a lot of room to be turned down.

So, going beyond what almost anyone would consider reasonable, time to start fabricating a transition piece for a big fan. 



This is the aluminum chassis of the power supply (on the right) and a mounting flange for the 120 mm fan being held in place by a block of wood.  Cereal box cardboard and masking tape are being used to make a template of a transition piece.  Unnecessary bits of the power supply chassis have been cut away already.  Here's the resulting template, ready to transfer to sheet metal:



Looking for suitable sheetmetal, I found something in an e-waste bin:



It just seemed right...

This view is the back of that bulkhead that holds the power supply.  Some of it was cut away for the fan and transition piece.  I went with the Noctua F12 PWM fan, which is the beige thing with the brown corners.  The blue stuff is RTV engine gasket compound, which was on-hand   It is a silicone sealant, used here to seal the gaps in the sheetmetal work.



The power supply was shifted about 35 mm to the right to make it fit.  I also removed those three big caps and mounted them on the right (using 600V rated wire).  One of the three mounting screws for the floppy drive was sacrificed, but it's still well attached (and I don't have anything else that uses floppies anyway).  Here's the side view:



(The grey band to the left of the fan is the ribbon cable to the CD-ROM.  It's not helping to promote good cooling, either...)

I took one of the thermistors that came with the fan speed controller and cut off its metal shroud.  It then fit snugly between the 5V and 3.3V modules in the power supply, near their switching transistors.  I figure this would be about the warmest place.   This is set as fan 2 on the controller.  With five cards in the unit, and now drawing about 690 W, the temperature reading from the power supply is 38.0C.  This checks with how PS chassis feels, as well as the air coming out.  The fan is running at 1080 RPM.  Maximum speed of the fan is 1500 RPM, so there's room for the controller to run the fan faster. 

And the noise?  I've just got a cheap tablet with a noise meter app.  Before any work, 30 cm directly in front of the unit it read:

Ambient level: 23 dBA average; with unit on: 37 dBA average (although it sounded like a lot more...)

After the steps described here:

Ambient level: 24 dBA average; with unit on: 25 dBA average

I'm sure it is considerably louder than this app says, but there is definite progress.  Now it's closer to lawnmower than hovercraft.    The hard drive is now the biggest noise source.  I've ordered a SCSI2SD board to take care of it.

There's more checking and tweaking to do.   One thing is that the cards draw a lot power.  If unused cards are replaced with filler panels, the  fan speed controller can turn down power supply fan speed, saving some noise there. 

Anyway, I hope this might give some ideas for options to quiet noisy test equipment fans.

Cheers,

Tim

[HalFET]

Impressive difference 

The scary part is, these are quiet versus the PCI acquisition cards Agilent used to make.    There they had two anemic fans running at their limit to cool the FPGAs/CPLDs.

[alm]

I like your systematic approach to lowering the noise, rather than replacing the fans with some low speed, low static pressure fans with sleeve-bearings and calling it a success if no smoke comes pouring out within a few hours of running idle at low ambient temperature . I obviously question the accuracy of the noise measurements, especially because measuring at shorter distances than the typical 1 meter can severely affect the measurements. But measuring at longer distances obviously requires more sophisticated equipment and lower ambient noise (isolation).

I can confirm that the units are pretty loud. I guess the engineers spent no further effort on optimizing airflow because they were within their thermal limits in the specified ambient conditions, and because noise was not much of an issue back then. Even office PCs were pretty loud until maybe 10-15 years ago.

[abraxa]

Maybe it's a good idea to borrow a thermal camera from a friend to make sure no ICs are uncooled that the chassis fan was pushing air across before?

[TimInCanada]

Hi guys,

Somebody said that these LAs have such great networking capabilities, because nobody wanted to be in the same room as one.    

I expect noise wasn't much of a marketing consideration when these were designed.  Customers were companies, and the analyzers would have been used in rooms with other equipment also sporting noisy fans.  A quiet machine would probably have had zero marketing advantage.  The man-months of engineering work that would have been needed to design more effective cooling would have just had no economic payback.

Us users, on the other hand, can put some thought into quieter cooling.  We don't have to validate our changes for 50C ambient operating, or for mechanical shock or vibration requirements, or for manufacturing tooling and other costs.  Sometimes you can get big improvements at little effort with a little thought.

A thermal camera would be very helpful, especially to quantify temperatures and measure the effects of cooling changes.  Unfortunately, I don't have access to one.  The motherboard can be felt by hand while the unit is running, so there is confidence it doesn't have any hot spots.  The power supply and cards in the card cage are more of a challenge because of physical access for either a thermal camera or feeling by hand.  I can feel some of the power supply outer surface and the temperature of the cooling air coming out and infer it is being cooled acceptably, assuming the power supply is designed with adequate heat sinks internally.  For the cards I have to assume that any hot spots would still be detectable in the time from unit shutdown until I can get a card out of the cage.  So far I haven't found anything too warm to comfortably hold onto.  That said, I think there are still some improvements that can be made.

The noise measurements I did are certainly mickey-mouse.  Qualitatively, the unit is much better now.  The hard drive has a high pitched sound component that is not loud in absolute terms, but I find it irritating over time.  It's an older SCSI drive and I don't know if any compatible replacement would be any quieter, so will try the SCSI2SD board.  An SD card has a limited number of write cycles, but that shouldn't be too much of an issue for a LA.

May Santa Clause be good to all the good little engineers, 

Tim

[texaspyro]

It's an older SCSI drive and I don't know if any compatible replacement would be any quieter, so will try the SCSI2SD board.  An SD card has a limited number of write cycles, but that shouldn't be too much of an issue for a LA.

That might be a bit of a problem...  some of these old devices have a limit on the max drive size so you may need to find a smaller capacity SD card.

Also you need to use industrial grade SD cards.  They are rated for a lot more write cycles.

I do sell an SD card replacement for the HP16500 analyzers hard drives ($40 loaded with the software).  It was quite a problem finding an SD card that would work with the 16500's.  The 16500 uses some obscure IDE commands that most drives / cards do not support.  I thought I found one that worked well so I bought all the cards the place had (at a stupid expensive price for 2 GB cards).  It turns out that identically marked cards used one of two different controller chips and less than half of them were compatible to the 16500.

[MauriceS]

As my SCSI CDROM was dead, i ended up replacing it with a SCSI2SD v6 with the CDROM image as well as the harddisk. That way i could remove the very long SCSI cable as well, and replace it with one having only 3 connectors. The SCSI2SD is now neatly sitting on the old harddisk bracket on plastic standoffs, rotated and power is taken from the term power. I used a higher grade 32GB SDHC card, 16G HDD, and 800MB SCSI CDROM emulation drive.

CDROM is no longer installed, but the external connector should still work. Ignite went really fast...

[DaJMasta]

I've done fan swaps on many of my units... but replacements can be tricky.  I've been hesitant so far to mill out or otherwise remove the inbuilt fan gratings - yes, they do create a lot of turbulence and noise, but I'm not sure I want a hole in a big shielded chassis of that size, so I haven't bothered.  If there's the clearance, you can use a couple washers or other spacers to hold the fan slightly away from the grating - goes a long way towards quieting down noise from close obstructions.

I've been usually using Scythe fans to replace high airflow main chassis fans, their 92mm PWM controlled fan and one of their 120mm fans offer about the same CFM as some of the noisy 120mms used in scopes and such, albeit at lower static pressure, but since that sort of application is generally fairly unobstructed, the static pressure counts for less.  Trickiest replacement type I've found is the same sort of very dense power supplies - they're typically sub 80mm sized, requiring high RPMs for decent airflow, and since they're so densely packed, the static pressure of the replacement really needs to be a match or near it.  Also ran into an issue with a power supply module in an Advantest analyzer that had a fan monitoring pin that would be communicated to the mainboard and auto-shutdown the system.... but it wasn't a standard PWM monitor but a simple on-off signal from the fan, which meant finding a replacement fan was tricky unless I wanted to either ignore the RPM warning or bodge in a circuit to detect the tachometer signal and converted it to the pass/fail it was looking for.

Another big problem for fan noise is CPU coolers in these things, cause they're often small, high RPM fans that are built into the heatsink instead of with a standard housing and holes.... but there tends to be more space in the CPU module, so I often replace the heatsink and strap on a standard fan or a larger blower fan undervolted on the side of the heatsink.

[MauriceS]

Has anyone changed the left rear side fan? As shown in the picture I attached for replacing the SCSI drive with a SCSI2SD drive (and thus removing the air blockage of the flatcable going to the CDROM - so in fact I improved airflow there) the fan is mounted using screws from the inside of the fan, whereas the Agilent service manual shows otherwise.  https://literature.cdn.keysight.com/litweb/pdf/16700-97015.pdf?id=574181 (I couldn't find a direct link on the keysight site - but google finds it)
I'd like to replace it and the center ones with PWM fans, either Noctua or Scythe, but have no clue how to? Do i need to tear this unit all the way down?