Catalog Filters

I was working on a project for the radio club and needed a couple of filters. One for 571 Mhz and the other for 735 Mhz. Previously I made a stripline filter and I have done copper pipe filters. These are rather large and I wanted something quick. Looking in the Digikey catalog I found filters by Toko. If you are buying a million of these filters I am sure Toko would make them exactly to your spec. For ham use in single quantity we have to take what is stocked in the catalog. Of course the catalog selection was made without consulting us hams to what we might need. The chance of finding something in stock value that is useful approches Epsilon. Much to my surprise I found a helical filter for 734 MHz with a 10 MHz bandwidth. Ok, lucked out on that one. For 571 the closest I could get was 550 MHz. Could I tune it up to 571? This little two section filter has tuning screws at the top of the cans. The screws were near the top so it looked like I was already at the top of the frequncy limit. As it turned out I was able to tune up to 571 MHz. For a test I ran the screws down to see how low I could get and still have the same overall filter response. The stock specification was Fo = 550 MHz, 10 MHz bandwidth. I don't recall the published insertion loss but I was looking at about 5DB. That was measured at the dip of the ripple which was approximatly 2 DB. My test jig was not ideal for grounding the cans, terminating the filter, and making connection to the RG188 teflon cable so I am sure I introduced some error in the response. The results:

Minimum Fo = 494.1 MHz
Maximum Fo = 586.0 Mhz

Other specs remained about the same but the insertion loss and bandwidth was increasing as I approched the upper limit.

For this filter I can pull about 8% from center but the catalog frequency was already near the top of the range. This would be unknown to the purchaser until after they examined the product.

If you need a filter check the catalog offerings from Digikey and Mouser. You may luck out and find something that might tune to the desired frequency but I would not try to go beyond 8 to 10 percent.

For reference this part was Toko type 7HW, Part number # 252HXPK-2736F (Digikey #TK3307-ND)

Followup on Cheap Cables

This is a follow up of the cable problem of the previous post.

From the photograph you can see there is minimal copper braid but an inner foil around the center dielectric. Looking into the cut-a-way connector body I found the crimp was applied metal against cable jacket. There were strands of the copper braid sandwiched between the metal connector body and the cable jacket. That was the extent of the ground connection. The foil did not touch the connector body as the inner chamber of the connector widened out past the cable jacket. I suspect the shield connection had been compromised. The bad connection could have been caused a number of factors such as the copper braid being pushed into the jacket, corrosion of the copper wires, or breakage of the wires.



Moral of the story is if you obtain cheap cables they may work fine but keep in mind they may have limited life. Don't be afraid to chuck them at first sign of trouble. I still prefer my homemade cables made from LMR-240 Ultraflex and the correct connectors for that cable type.

Cheap Cables

I picked up a box of 3-foot and 6-foot BNC cables. The cables are marked as RG58. Overall construction of the cables look good as they have molded ends on the BNC connectors. Unfortunately good looks is where it ends. Electrically the cables suck. A good example raised its ugly head today. I was checking a function generator and connected it to my HP 5335 frequency counter. Ran through a few frequencies to make sure the calibration of the generator was reasonable. Higher frequencies no problem. When I went to 1 KHz the counter would not lock onto the signal. I connected the generator to my oscilloscope and observed a nice 1 KHz sine wave. Playing around with the counter I engaged the filter switch and got a correct measurement of 1 KHz. Looks like I am getting noise in the input on lower frequencies that prevents the counter from locking onto the desired signal. I placed a few clamp-on ferrite chokes on the cable and that also took care of the problem. I then replaced the cheap BNC cable with one I made from LMR240 ultra flex. Problem went away without any filtering.

Moral of the story. Buy good cables they will save you headaches in the long run.

ABS problems on the van

I am having ABS issues on my van. The ABS brake system kicks in at very low speeds on dry pavement. There is a service bulletin about rust on the mount point of the sensor causing it to pull away from the hub so the signal is erratic. That would be a simple fix as the sensor is $200. I can not find a cheap overseas replacement part even on Ebay. My scan tools shows a couple of generic ABS chassis codes:

C0281 (chassis code - ABS) Brake Switch Circuit
C0223 (chassis code - ABS) Rt Front Speed Signal Erratic

I suspected the right front wheel so this confirms it. The test procedures specify resistance and voltage readings for the sensor. Getting the voltage reading is going to be a trick as I can not spin the front wheel as I have all wheel drive. I will have to disable the ABS then rig up a cable to attach to the harness so I can roll the van down the driveway to measure the AC pulse train. The voltage is very low (approx 100 millivolts) so noise is going to be a problem even using the oscilloscope. The service bulletin indicates the signal should be at least 350 mV but this document says greater than 100 mV. So much for consistent documentation.

The procedure also measures the cable from the sensor to the electronic brake control module (ECBM). It states that faulty wire connections is a leading cause. The darn EBCM is attached to the frame under the van. That is the dirtiest place you could put the thing.

I am a little confused by the C0281 code for the brake switch circuit. The description indicates that this code may set if the there is an ABS fault of at least 500 milliseconds. That would have set by the C0223 code but then in the next line it says the code only sets if there is an open or short in the brake switch circuit wiring. For now I will assume it is setting because of the sensor signal. Fix that and then see if C0281 comes back.

The last item in the correction chart is to replace the ECBM. Unfortunately you have to have a scan tool that communicates to the device. Consumer scan tools only read diagnostic codes. You need on of the expensive units to talk to the devices. You have to set the tire diameter for the vehicle in the ECBM. Crap! If that is bad it is back to the dealer to replace that darn thing. I bet they get a lot of $$$ for that module.

Over the weekend I hope to get a chance to investigate the sensor/wiring.

UPS Batteries

Test complted on the batteries in the UPS.  There are four PowerSonic PS-1272 7.2Ah batteries in the UPS.  Three of the pass as shown in this graph:

 

 

One battery would not carry any load.  That one is the problem. 

 

Results of the test in text format:

 

 

 

UPS Failure!

I have a 1400 watt rack mounted UPS that powers my computers. The computers and UPS are in a rack. Yesterday I discovered all my computers rebooted in the early morning. The UPS appeared to be running fine by way of the indicator lights on the front. I gave it the big test by pulling the power cord from the wall. The entire rack went dark. The UPS was running in bypass apparently. I brought the UPS back on line and now I had a red malfunction light on the front panel. This is a refurbished UPS which is one of two that I have. They are about a year old. I swapped the failed unit for the other one which is only running one PC downstairs. I am now in process of checking the batteries. I have noticed the battery charge status leds have been flashing. I have a technical bulletin about that indication. The document talks about a calibration process that may remove that warning. Before I do that I need to test the batteries to make sure they are still in spec. There are four batteries in the UPS. They are 12 volt Power Sonic 7.2AH batteries. They are connected in series and parallel providing 24 volts at 14 Ah. I have about 650 watts on this UPS so that only gives me about 10 to 15 minutes run time before they crap out.


I test the batteries using a computerized test fixture. This is a device by West Mountain Radio. Plugs into the USB port on the PC and by running the provided software you can do a load test and see if the battery meets specification. You can see the resulting chart after the test on the first battery. You set the load current as indicated by the specification. Most batteries of this type are tested over a 10 or 20 hour rate. Power Sonic lists the spec for both 10 and 20 hours so I used the 10 hour rate. For a 7.2 AH battery that is 0.700 Amps. They also specify the "dead" voltage which is typically 10.5 volts for a 12 volt battery at this load current. As you can see from the chart I got 8 Ah out of the battery by the time it hit 10.5 volts. So battery 1  is good. This battery goes on the charger and I connect up the second. At ten hours a test this will take a little time.


Chart after test of first battery



Batteries ready for testing. Battery #1 connected to the test fixture



Batteries as wired up in the UPS unit

New Toy

I picked up an HP 5372A Frequency and Time Interval Analyzer. This box take time interval measurements: time interval, continuous time interval, and +/- time interval. A time interval is a measurement of elapsed time between two electrical pulses. I will use it for testing and comparing oscillators. As I sit here playing with this unit I wonder what kind of unusual time interval analysis has been done using this type of device in industry or the military.


I have completed the functional tests out of the manual. The unit appears to be in good shape and calibration.


 

The SUN's solar activity and climate change

All this crap about man being the cause of global warming never ceases. In the political arena man is to blame and everyone totally discounts the effects of the SUN. What a bunch of moron’s using nature’s normal cycle as an excuse to generate taxes (cap and trade/carbon emission reduction) and force lifestyle changes on the population. Years ago I took a course on the Sun’s effects on radio propagation. This was long before Owl Gore and his CO2 vs temperature chart (which he read backwards by the way). It was quite apparent that the 11 year solar sunspot cycle’s overall trend has the greatest influence on the Earth’s temperature. Much more than what man can do to the planet. We have been at the bottom of the last solar cycle for a much greater period of time then cycles in the past. If the new solar cycle had followed the pattern of previous cycles we would now be a couple of years into a rising sunspot count. This pattern has not occurred in this cycle. We have been in a period of no sunspot activity. We have been in the low period of sunspots for a few years. The comment below speaks of the period of time where NO sunspots have appeared at all. Continued low solar activity will drive the Earth temperatures down. In the current political arena the wording is changing from “global warming” to “climate change”. That way they cover all bases no matter what the temperature does. They can continue to spread lies about climate change to effect new taxes and control of the population.


This is comment was in a DX news letter that follows along with what I found when I took the propagation course.


Out in space the sun remains on a record pace at 51 days without sunspots, and no expectations of any change soon.  We’ll be in the top 3 longest quiet periods by Wednesday evening (tied with 54 days from 1879) and by late next week we’ll head into the 2nd position as we surpass 63 days from 1901. The daily solar observations began in 1849, but we know from other observations between 1400 and 1830 that these
periods of sunspot minimums were associated with a 400-year stretch known as the Little Ice Age. The only problem with observations from that time is that there were no telescopes and satellites with the observational power we have today. In other words, the long periods without sunspots are even more impressive now because we can pick out the smallest sunspot, ones that would have been missed 100 to 200 years ago. We
won’t know if this is true for a long time, but at this point there is no reason to discount the idea that we’re heading into another Little Ice Age, or possible a major ice age. Time will tell…

Testing ERA-5 MMIC for use as a multiplier

An easy way to generate harmonics is to drive an amplifier into compression. MMIC amps are easy to use and the 1 DB compression point is listed in the datasheet for just about all MMIC amplifiers available today. Normally this would be an undesired operating point. If you need a multiplier this a simple way to do it. Output are typically higher using an MMIC as a multiplier than using a simple diode multiplier. The output circuit of the MMIC can be designed to enhance the desired harmonic. That was not done in this test.

ERA-5 gain below 1GHZ = 20 DB
1 DB compression = +17.5 DBm

Input 738 MHz into MMIC at +6 DBm and +2 DBm drives the amp well into compression. No attempt was made to optimize the output for a particular harmonic or to suppress the fundamental. The output capacitor was 12 pf to favor the upper frequencies. Results varied depending on the input frequency. The data below is for 738 MHz. At 571 MHz the 5th harmonic with +2 DBm input was -20 but had a lot of “grass” around the carrier. Lowering the input to +1 DBm dropped the harmonic to -30 DBm and cleaned up the noise (or it dropped below the noise floor of the spectrum analyzer).

It appears the input should not exceed 5 DB above the 1 DB compression level or excessive noise will result.

+6 DBm input is 8.5 DB above the 1DB compression point
+2 DBm input is 4.5 DB above the 1DB compression point

Output 1 is with a +6 DBm input
Output 2 is with a +2 DBm input

Frequency(MHz) or Harmonic/Output1/Output2
738/+20/+18
X2/-2/-6
X3/+3/-12
X4/-4/-30
X5/-20/-30
X6/-42/-34
X7/-38/-46
X8/-30/-42
X9/-40/-50
X10/(below noise)/(below noise)

Oscillators

I assembled the two transponder oscillators yesterday. One is for 82 MHz and the other for 114.2 mHz. They are based on the initial design I did last summer. I thought I had these all figured out. The 82 MHz oscillator just sucked. The harmonic content was very high. Hell, I did not need the multiplier stages I designed. I had a 738 MHz harmonic from the oscillator directly only 30 DB below the fundamental!! As it turned out the little MMIC amp I had at the end of the oscillator was being overdriven. I just removed the MMIC and jumpered the board from the end of the buffer to the SMA connector. I have a good -5 to 0 DBm which was the initial design goal. I also removed the filter components after the buffer. I was getting a spur around 300 MHz. I don't know if it would have gone away by removing the MMIC. I removed the filter parts before removing the MMIC. I suspect circuit board traces in the filter caused the instability. I probably had stray capacitance and/or inductance somewhere in the filter physical layout that screwed it all up.

Other changes I made:
1. Removed the tuned transformer between the oscillator and buffer. Replaced it with a short to the DC line.
2. I moved the input capacitor of the buffer that was on the transformer link to the top of the oscillator emitter resistor.
3. I also reduced the buffer emitter resistor from 220 to 100 ohms. This helped reduce the harmonics a little. I think the buffer was also being overdriven.

I experimented with the tapped capacitors in the oscillator tank circuit. Increased them to 82 and 33 from 68 and 22 pf. No real difference. The oscillator may be slightly more stable, from a start up perspective, with the original caps so I put them back to the way they were.

In the end the 2nd harmonic is 10 db down and the others are 30 db down or more. The 5th harmonic is the last one I see on the spectrum analyzer. The input tuning on the multiplier may make these spurs a moot point. If not I will have to add an extra low pass filter after the oscillator. The final oscillator is much simpler design.

I will incorporate the design changes into the 114.2 MHz oscillator and test it out tomorrow night.

I guess I will have to go back and update the design document on my web page. That will have to wait until after Hamvention. I need to get this project done by Hamvention.

Interdigital Filters - remade

I etched the new boards over the weekend. The frequency of the 738 and 571 filters came out spot on. The pesky "step" is still there especially on the 571 filter. The loss is less on the 571 but that was expected per the simulation using FR4 board stock. The return loss just sucked on both. I ended up adding a surface mount trimmer cap (3 to 12 pf I think) in series with the input on the 571 filter. I also moved the input tap up several millimeters. About as far as I could reach with the body of the trimmer. Adjusting the cap I was able to achieve a return loss of 18 DB. Slightly above the desired frequency I could get 20. I adjusted for best RL on the design frequency. When I get some time I will add the same capacitor to the 738 MHz filter. For now these are done and I will go on to other modules of the project.

Making New Artwork for 738 Interdigital Filter

One final mucking with Ansoft dimensions on the interdigital filter. I started by changing the microstrip impedance to 82 ohms (40 mils on this FR4 board). It was around 40 or 50 before from the initial Ansoft design. That appears to work much better. On the pipe cap filters the ideal impedance of the pipe coax structure was 77 ohms so that must hold in general for the micro strip filters too. The filter response is much better with the best return loss also being the center of the band pass response. The RL looks to be 15 or so. Insertion loss is still high but down under 15 DB now. Out of time for tonight so I will re-do the PC board artwork in Eagle tomorrow night.

Update on Frequency Error in Ansoft Design

I found the source of the 100 MHz frequency error in the earlier post. Ansoft converts the theoretical filter design into a physical layout. This layout is more of a schematic then a circuit board layout. The four microstrips of the filter are modeled in Ansoft as four rectangles. Three sets of these rectangles make up the entire filter. Each strip is broken up into three pieces because of the input and output taps. The taps are represented by a “T” symbol that has no length, just width. To position the tap the total length of the line must be broken up in the model representation. There is some confusion as to what length “p” is in relation to the total length of the line. Also the conversion process did not use the variable “p” but instead inserted the total length of the line. When I made the board I misinterpreted the meaning of each length “p”, i.e. p1, p2 and p3. It was not clear how these added up to make the total length. I ended up with each filter element being about 250 mils too long. This made the filter resonate 100 MHz low. Now that I have figured this out the simulated frequency and measured frequency is within reason give just an uncertain dielectric constant of FR4 at these frequencies.

Measurements on the N4 Interdigital Filter

After messing around with this interdigital filter I never could get a nice smooth response. At best there is “step” in the response as the frequency goes above the center frequency of the pass band. For this application that would not be a problem as I am just removing other harmonics of the main oscillator. Using FR4 board is not the best for UHF and above filters. The loss is just too high. Also the dielectric constant is not very constant. Now that I have this filter built I measured the final microstrip elements.

P (total length of each microstrip, there are four) = 2062.5 mils
P2 (tap for the input and output from ground end) = 203 mils
W (width of each microstrip, they are all the same) = 164 mils
S1 (spacing between element 1 and 2) = 391 mils
S2 (spacing between element 2 and 3) = 172 mils
S3 (spacing between element 3 and 4) = 500 mils

Measuring the actual center of the band pass on the tracking generator indicates 737.68 MHz. Now we can modify the simulation with our actual measured values and keep changing the dielectric constant until the frequency matches the measured. Nominally FR4 is around 4.5. The simulation shows:

For 4.7 Fo = 743.53 MHz
For 4.8 Fo = 736.39 MHz
For 4.9 Fo = 728.57 MHz
For 4.79 Fo = 737.68 MHz

Setting the simulation to 4.79 yields our measured frequency of 737.68. We will use this value for future filters around this frequency. I want to try a hairpin filter next. I have the initial design. Other than the narrow spacing and I assume more critical layout tolerances the pass band does not contain the “step”. Hairpins do respond to the even harmonics of the first pass band. I don’t think that will be a problem the higher harmonics of the multiplier will be reduced by the design of the multiplier.

Documenting the shortening of the MicroStrips

Start length = 61mm

 

Pass band frequency per length of each microstrip:

61mm = 630 MHz

60mm = 642.5 MHz

59mm = 657 MHz

57mm = 678 MHz

54mm = 715 MHz

53mm = 728 MHz

52mm = 736.7 MHz

 

 

After Microstrip Adjustments

I have made the modifications to the 738 filter PC board layout.  I shortened the strips to bring the center frequency to 738 MHz. The original length from the Ansoft calculations was 61mm.  I had to shorten these to 52mm.  That is a reduction of 14.75%.  My guess the reduction is required because of stray capacitance on the board.   The final 3DB bandwidth comes out to 15.1 MHz.  The final gain with the +20 DB MMIC is 0DB.  My insertion loss is 20DB behind that amp.  This is in line with Ansoft’s calculations using FR4 board.  Rogers PC board would be a better choice because of the dielectric loss tangent (TAND).  It is 0.02 on FR4 but only 0.0009 for Duroid 5880.

 

The input return loss is interesting.  Using FR4’s TAND value Ansoft shows the best return loss as only 10DB but above Fo at 756 MHz.  This is less than the original 18DB I had before mucking with the strips.  I see this on my tracking generator.  I was able to improve this to 15DB by adjusting the tap point on the first microstrip element.  I moved it a millimeter or two toward ground.  I don’t know that it improved return loss at 738 as that still looks to be 10DB. 

 

 

 

738 MHz Bandpass Microstrip tuning

I was playing around with Ansoft this morning and the 738 filter. First I adjusted the dielectric constant to see if I could account for the lower filter frequency response.  I found the dielectric constant would have to be something like 7+ to drop the frequency down 100 MHz.  Design spec for FR4 at lower frequencies is 4.5.  I don’t think 7 is a reasonable value even at 700 MHz.   I then checked my actual dimensions of the micro strip.  They are within 0.5 millimeter of the design values.  Again, not enough to account for 100 MHz.  I am beginning to wonder if stray capacitance may be lowering the frequency. 

 

Another finding was on the insertion loss.  When I modeled this board I did not set the dielectric loss tangent value.  I see from documentation that this should be 0.02 for FR4 board.  Doing that I can now account for the loss.  The insertion loss is 18 dog biscuits (Jake is pawing me as he wants some of those).  That agrees with my board.  My MMIC amp on the output has a +20 gain at 700 MHz.  That gives me the +2 DB output with a 0-DBm input signal I see on the analyzer.   Using Rogers 5880 the TAND values is 0.0009 which would produce a more reasonable loss of 4 dog biscuits and improve the return loss.   

 

Next I will shorten the strips to see if I can bring the frequency up to 738.  I can try to use the same technique I used for the pipe filters to tune each stage by lightly coupling a spectrum analyzer to the first stage then short out the other stages as I tune down the line.

 

 

First Microstrip Bandpass Filter Results

I etched the board for the microstrip band pass filter.  The desired center frequency is 738 MHz.  The problem with designing such a filter on FR4 PC board stock is the variation of the dielectric constant at the desired frequency.  The second source of errors in the layout of the board itself.  At UHF and above it is difficult to get the necessary precision where a quarter of a millimeter could make a big difference.  My results of the first filter look promising.  The filter response was about what I wanted except the frequency was off by 100 MHz.  The center of the bandpass was 630 MHz and not 738.  I think the insertion loss may be high but it is masked by the output MMIC amp I have on the board.  The overall gain is +2 DB because of the amplifier.  While this does not sound bad remember that the MSA-0886 amp has about 20 DB of gain at 600 MHz. 

 

The plot of the filter response shows a “step” on the upper frequency side of the curve.  This indicates one of the stages is not aligned on the same frequency as the others. 

 

Anyway, the initial measurements:

 

Input return loss at center of bandpass:  18 DB

Overall gain with MMIC amp: +2 DB

Center frequency of band pass:  630.68 MHz

3db bandwidth:  638.33 – 623.64 MHz = 14.7 MHz

 

The filter was designed using Ansoft Designer SV (student version). I will now go back to the design and modify the FR4 dielectric constant to see if I can account for the 100 MHz frequency error.  The idea is to use the modified constant to rework the board dimensions to see if I can get closer to 738 MHz and improve the insertion loss. 

 

As for this board it has served it purpose.  I can try to shorten the microstrips and use capacitors to tune the sections.  I can not adjust the coupling other than to solder metal tabs onto the microstrips and use them to increase coupling to the next strip. 

 

 

 

 

Tinkering With That Darn Oscillator

It appears the high harmonic output is generated by the last MMIC amp.  Not sure why unless it is sensitive to the impedance seen by the input looking back into the filter.  There is a simple low pass filter after the oscillator buffer.  That filter should have been after the MMIC amp but the amp was an after thought when I designed the PC board.  The harmonics go away when I remove the last output capacitor from the filter.  I then have the second harmonic down 35 DB.  I attempted to replace the first and last capacitor with trimmers to see if I could improve the situation.  Not really, so I left it as it was minus the last capacitor. 

 

I adjusted the oscillator transistor bias resistors to improve the output while maintaining the same 35 db harmonic attenuation.  The oscillator output level is now zero DBm.

 

That still puzzles me as to why I am getting the harmonics.   The MMIC is not being overdriven.  The MMIC is an Agilent ABA-53563 I got from Mouser Electronics.  The 1-DB compression point is +14 DBm output.  I appear to have around -15 DBm going into the MMIC so with a 21.5 DB gain that is below the 1-DB compression point.  I checked my spectrum analyzer calibration against my signal generator.  The spectrum analyzer reading agree with the settings of the generator. 

 

I will leave the oscillator as it is and get back to testing the frequency calibration and oscillator heater.  Something tells me I will be redesigning the PC board moving the filter after the MMIC. 

Oscillator Oven

I have been working on a few oscillators for a radio project.  The test oscillator is crystal controlled on a carrier frequency of 104.727270 MHz.  This will be multiplied up to a higher frequency.  Unfortunately I forgot what oven temperature I ordered the crystal for.  I think it was around 50C.  I did not want it to be too high as I am using Styrofoam 3/4” thick for the insulating material. The entire oscillator is heated.  The oscillator is a Butler design which uses a series resonant crystal.  The only tuning is from the resonant circuit of the Butler design.  I set the trimmer capacitor for the best oscillation then adjusted the oven heater until I moved it to the marked crystal frequency.   The oscillator has a temperature sensor on it so I can monitor the temperature.  This makes adjusting the heater easier.  The sensor outputs 0.01 volts/degree K. 

 

I now have a stable operation at a sensor reading of 3.292 V which equates to around 56C.  The frequency counter reads 104.727284 MHz.  The counter is locked to 10-Mhz GPS disciplined oscillator to ensure a correct readout.  I am 14-hz off frequency at this point.  The crystal changes about 57-Hz for a 2 degree C temperature change.  This is around 0.54 PPM.  The oven heater resting current is 0.06 Amps.  The heater is nothing more than a power FET attached to the side of the aluminum oscillator housing.  A thermistor on the oscillator PC board in the gate circuit controls the FET.  An external pot in series with the thermistor is used to adjust the temperature.

 

Now I will remember to write down the temperature specification when I order any new crystals.  55 degrees C. works out well for the type of heater and insulating material I am using. 

 

The next test will be to see how well the heater can hold this temperature when I subject the oscillator package to a temperature change.  I need measure the external and internal temperature while the environment changes.  I can do this by using the GPIB interface on my DVM for the on-board sensor and the RS-232 interface on my Radio Shack hand held DVM.  The sensor is a LM324.  I hope to be able to hold +/- 50 hz from room temperature to an ice bath. That would be 2 degrees C temperature variation.