Low noise low distortion sine wave generator

[bobwidlar]

Hi,

Recently I've been working on the test setup for my 20-bit audio ADC (I'm waiting for the chips to return back in 2 months). It's really hard to find an affordable signal generator to find with the specs: lower than -104 dB THD and < 10nV/sqrt(hz) noise floor. A floating ground option would be nice.

The easiest to have an audio analyzer from AudioPrecision, but I don't need analog or digital signal acquisition part of it. So, it's kind of waste of money.

I've found Shibasoku AG15C has the best distortion performance. But no clue on noise floor. I send them e-mail after e-mail asking a proper datasheet, they kept sending some documents which does not give any proper information at all. I can't imagine how the support will be after purchase.

There's also Stanford Research Sys. DS360 http://www.thinksrs.com/products/DS360.htm . It's good for its price, but the noise floor is incredibly high. 150nV/sqrt(hz). I will need some serious filtering and attenuation it seems.

So, do you know any alternatives?

[leppie]

It's good for its price, but the noise floor is incredibly high. 150nV/sqrt(hz).

But that is for the 1.26 Vpp to 40 Vpp range. I dont know much about audio or what you doing, but that sounds like a lot!

[bobwidlar]

Hi, my ADC has 3.6 Vpp input range. So, I need to go > 1.26 Vpp range which is really unfortunate that it amplifies the noise by 10 in that range. So, I need to use the maximum output swing and attenuate it by 16 times or sth to get required SNR 

[robrenz]

This thread may help

[bobwidlar]

It's a great topic. I've already read it but I'm not willing to build my own source. Apparently the signal generator there isn't available as a ready kit or assembled PCB anymore.

[alexwhittemore]

You can afford to spin an IC, but you can't afford the equipment to test it? I'm just terribly confused.

[bobwidlar]

Well, this is about being a PhD student. Tape-out is for free, but meas. instruments should be bought by the university. Nobody (supervisors, hehe) wants to invest 20k to buy an Audio Precision 2722 for an one-time use if it's avoidable. If you think to it, you have a given budget for whole of your thesis. It's all about how you spend the money. I will buy some hand-made sharp filters for quite some money for example.

Well, I've found a DS360 in the lab and playing with it already. If anybody is curious about the instrument, just let me know.

[bobwidlar]

Argh, forgot to say. We have an option to borrow AP-2722 from another lab. I guess now it makes more sense to not buy that.

[T3sl4co1l]

Note that, if you're recording a sine wave of known distortion (ideally, none -- you can use a high order, low distortion lowpass or bandpass filter to assist in this), noise can be integrated out because it is incoherent and averages towards zero, while the distortion (which is phase locked to the fundamental) remains coherent.

You can also try testing it with a very linear ramp, which can be generated with a very stable square wave and an integrator (precision op-amp with high quality feedback capacitance -- polypropylene, polystyrene, teflon, or better still).  Obviously, a least-squares fit to the resulting data series easily finds errors (INL/DNL), and the line equation can be compared against the input to find gain and offset error (but now you need calibration).

The frequency used for the ramp will still matter, because inevitably there will be dielectric absorption and loss in the capacitor, which will skew the slope ever so slightly.  It should be as fast as possible (to approximate absorption as nil), but also as slow as necessary (ideally, hitting every code on the way up and down, thus proving a one-to-one transfer function with no missing codes -- assuming your design is intended to achieve this result).

Data arbitrarily close to the peaks (direction changes) should be discarded, due to finite rise/fall time of the square wave, ringing and settling, and finite frequency response of the op-amp and signal path.  (Ideally, a triangle wave has harmonics with amplitudes of 1/N^2, for N up to infinity; a real triangle wave will have some harmonics not exactly that amplitude, but higher or lower than others, and also phase shifted; and after some N, the amplitudes will start dropping much faster than 1/N^2, perhaps N^3 or N^4, before becoming nonexistent for all intents and purposes.)

Tim

[Jay_Diddy_B]

Hi,

You should read this: http://edn.com/electronics-blogs/anablog/4311362/Audio-Precision-to-the-rescue-of-Jim-Williams-no-less#comments

This is the details of the oscillator: http://cds.linear.com/docs/en/application-note/AN132f.pdf

And this document shows how to lock a Wein Bridge Oscillator by injection: http://staging.edn.com/design/analog/4400342/1/Injection-lock-a-Wien-bridge-oscillator

Jay_Diddy_B

[bobwidlar]

Note that, if you're recording a sine wave of known distortion (ideally, none -- you can use a high order, low distortion lowpass or bandpass filter to assist in this), noise can be integrated out because it is incoherent and averages towards zero, while the distortion (which is phase locked to the fundamental) remains coherent.

You can also try testing it with a very linear ramp, which can be generated with a very stable square wave and an integrator (precision op-amp with high quality feedback capacitance -- polypropylene, polystyrene, teflon, or better still).  Obviously, a least-squares fit to the resulting data series easily finds errors (INL/DNL), and the line equation can be compared against the input to find gain and offset error (but now you need calibration).

The frequency used for the ramp will still matter, because inevitably there will be dielectric absorption and loss in the capacitor, which will skew the slope ever so slightly.  It should be as fast as possible (to approximate absorption as nil), but also as slow as necessary (ideally, hitting every code on the way up and down, thus proving a one-to-one transfer function with no missing codes -- assuming your design is intended to achieve this result).

Data arbitrarily close to the peaks (direction changes) should be discarded, due to finite rise/fall time of the square wave, ringing and settling, and finite frequency response of the op-amp and signal path.  (Ideally, a triangle wave has harmonics with amplitudes of 1/N^2, for N up to infinity; a real triangle wave will have some harmonics not exactly that amplitude, but higher or lower than others, and also phase shifted; and after some N, the amplitudes will start dropping much faster than 1/N^2, perhaps N^3 or N^4, before becoming nonexistent for all intents and purposes.)

Tim

Hi Tim,

You're right about the test methods. With a signal which has a known amplitude distribution (ramp, triangular wave, sine etc) you can make a histogram test to get DNL, and than integrating that you can get INL. I did this kind of measurements before for an 20-bit ADC, and I didn't even used a square wave + integrator. I generated the ramp by AD5791 20-bit 1-ppm INL DAC, which I first characterized its INL and DNL by a 8.5 digit multimeter. This worked really great.

However, in this case I'm more interested in dynamic performance. For sigma-delta ADCs, generally INL and DNL is not defined since they are static characteristics and sigma-delta doesn't have a LTI transfer function. More important is to see the noise floor, noise shaping (if there's any leakage of shaped noise into baseband), distortion etc. at different freqs and different amplitudes. The most classical plot is input power vs SNR/SNDR. So, these are the kind of things I would try to measure first.

Hi,

You should read this: http://edn.com/electronics-blogs/anablog/4311362/Audio-Precision-to-the-rescue-of-Jim-Williams-no-less#comments

This is the details of the oscillator: http://cds.linear.com/docs/en/application-note/AN132f.pdf

And this document shows how to lock a Wein Bridge Oscillator by injection: http://staging.edn.com/design/analog/4400342/1/Injection-lock-a-Wien-bridge-oscillator

Jay_Diddy_B

Hi Jay,

Thanks for the links. I read app note 132 before but didn't see the others. Thanks a lot!

[T3sl4co1l]

Ah.  You didn't mention what kind of ADC you're testing, and of course you want to tailor the tests for the subject.

...Actually, to be perfectly precise, you still didn't state what kind of ADC you're testing, though I'm guessing that's why you brought up S-D ADCs.

Tim

[alexwhittemore]

Well, this is about being a PhD student. Ahhh, that makes MUCH more sense now

[PChi]

There is some discussion about this on http://www.diyaudio.com/forums/equipment-tools/205304-low-distortion-audio-range-oscillator.html

[bobwidlar]

Ah.  You didn't mention what kind of ADC you're testing, and of course you want to tailor the tests for the subject.

...Actually, to be perfectly precise, you still didn't state what kind of ADC you're testing, though I'm guessing that's why you brought up S-D ADCs.

Tim

20-bit audio-band ADC. It gotta be SD. The simplest reason is the size sampling cap. Of course there are other reasons like element matching etc (although you can go up to 14-bit with SAR and pipeline). So, yes it is SD ADC. A novel kind though.

[macboy]

These are the opposite of spending $20k on an AP unit:
http://www.ebay.com/sch/mickevich/m.html
At only ~$50, maybe it's worth a try. You have access to an AP unit from anther lab, so you could at least use that to measure the distortion from one of these to see if it will cut it.

[bobwidlar]

I arranged to borrow the AP-2722. So, I will be using that. Maybe I can take some record on that. I'm not sure if anyone is interested in pcb design of the system, I might open a topic for that maybe?

@macboy: I'm rather suspicious on that circuit but surely the price's worth a try. I will update here if I buy that one. I'm planning to buy a high-end commercial ADC to compare also. Maybe PCM4220's eval board for 150$ only.

[bobwidlar]

An update on this. I have APx555, AP-2722, UPV and UPD. I'm so shocked by the APx555's poor frequency accuracy on low distortion mode. Yes, it is low distortion but if you want to have nice plots with coherent sampling just forget about it. 2722 is the best and UPV comes close.

One interesting observation also: UPV and UPD are better than Audio Precision stuff to drive switched-capacitor inputs (I mean directly driving the input of an ADC).

I have 103dB THD and 110dB SNR measured from the latest silicon btw. not A-weighted, pure flat thermal noise floor. Flicker noise corner is around 1Hz.