[time-nuts] Triangle Waves

Magnus Danielson magnus at rubidium.dyndns.org
Wed Feb 3 00:26:51 UTC 2010


Bruce Griffiths wrote:
> Bob Camp wrote:
>> Hi
>>
>> At least from the last time I tried it:
>>
>> If you use a sine wave input source, it's got to be an amazingly good 
>> 10 Hz sine wave. A normal audio generator will not produce a 10 Hz 
>> output with good enough short term stability / noise to give you 
>> useful data. Audio generators may be out there that will do the job, 
>> but I certainly don't have one, and have never come across one.
>>
>> Since the output of the mixer is basically a triangle wave, it makes 
>> sense to use that as your test source. A triangle wave also has the 
>> nice property that it's easy on the math. You don't have any 
>> approximation issues with the integers going into the DAC. That shoves 
>> the inevitable digital crud higher in frequency.
>>
>>    
> When both the RF and LO ports are saturated, the mixer output waveform 
> depends on how the IF port is terminated.
> The output is indeed approximately triangular with your IF port 
> termination method when both the RF and LO ports are saturated.

My experience says that it also depends on the relative phase... so it 
shifts between approx falling saw to approx tri to approx rising saw to 
approx falling saw over a 360 degree beat period. Kind of comforting to 
see this shift occurring slowly on the scope.

> With the IF port terminated in a capacitor when both RF and LO ports are 
> saturated the output waveform is quasi trapezoidal.
> When only the LO port is saturated the IF output is sinusoidal.
>> Another nice thing about a pure digital approach is that it provides a 
>> clean trigger for the "start" channel of the counter you are testing 
>> things with. You can even set up the DAC to put out square waves to 
>> see just how good various bits of the chain are. Tough to do that with 
>> anything other than another arbitrary function generator.
>>
>> I agree that the reference is going to be an issue and that a LED 
>> stack may be the way to go. No matter how you generate the test tone, 
>> power supply noise will be an issue.
>>
>> The output amplifier on the DAC is my biggest worry. I could go with a 
>> current out DAC and something like an OP-27.  That won't give me 
>> 1nV/Hz either, but it will at least be within shouting distance of 
>> it.  Sigma deltas might be a third option. I have no idea what their 
>> low frequency flicker noise looks like.
>>    
> Producing a high amplitude (eg 20V pp) output and attenuating it down to 
> say 2V pp or so typical of a mixer will significantly reduce the noise 
> due to the output amplifier.
>> So, other than the noise issue (which obviously needs to be analyzed / 
>> tested / pounded on) any other issues with the approach?
>>
>> --------
>>
>> At least from what I have seen in the past, level sensitivity on the 
>> inputs shows up pretty fast in the output "beat note" as you vary the 
>> input signals that are supposed to be saturating the mixer. If they 
>> are doing their job, a 2 db level change produces a very small change 
>> in the output. If you have something amiss in that department, you 
>> will see it pretty fast. On that I'm pretty much in agreement with 
>> Rubiola's stuff.
>>
>>    
> Yes but NIST used a saturated mixer and still found that the mixer phase 
> shift depended on how hard you drive the diodes.

This should not come as a big surprise, as diodes change their 
capacitance with applied voltage.

> Long term variations in isolation amplifier output due to temperature 
> variations may be significant.

I haven't heard of AGC being applied to this particular system, but 
their use on stabilizing mixer phase detector gain should be known. 
Allowing loop gain to change with input signal strength is not always a 
good idea.

>> Since I intend to mate the isolation amps up directly on the same 
>> board as the mixer, there is no real need for a 50 ohm interface 
>> between them. If the mixer looks like 18.26 ohms,  the amp output can 
>> be transformed to that level rather than 50 ohms. Everything is 
>> matched (over a 1/8" trace) and you don't burn up power in a bunch of 
>> resistors. How well that idea works - time will tell. It's easy to put 
>> the resistors in if it flunks out.

I've considered that type of arrangement... it could provide some 
opportunities not allowed by cabled interconnect. For reasonable 
frequencies much simpler transmission models can be used.

Cheers,
Magnus



More information about the time-nuts mailing list