[hpsdr] NPR of Mercury (was: Boat Anchors..PreAmps)

Fri Jun 8 00:11:39 PDT 2012

Hi Alex,

Today was an ideal day for lab work - raw, wet, rainy, 12°C. So I spent part of the day measuring NPR on the Perseus in various configurations. First, I measured NPR with the preselector out, and a -3 dB noise-loading bandwidth of 35 MHz for the 1940 kHz notch and 30 MHz for all other cases. I derived correction factors for a hypothetical 40 MHz noise-loading bandwidth (fs/2). The preamp and dither were off for all cases.

The following results were obtained:

1940 kHz notch: NPR = 63 dB observed/76 dB calculated. Correction factor = 12.5 dB.

3886 kHz notch: NPR = 68 dB observed/78 dB calculated. Correction factor = 10 dB.

5340 kHz notch: NPR = 66 dB observed/74.5 dB calculated. Correction factor = 8.5 dB.

7600 kHz notch: NPR = 68 dB observed/75 dB calculated. Correction factor = 7 dB.

All four calculated NPR values were reasonably close to the theoretical 74 dB maximum shown in ADI MT-005, Fig. 2, p.3. 

I then repeated the four test cases with the preselector in. I used the same noise-loading bandwidths as above, first without band limiting and then with the appropriate band-limiting filters selected in the noise generator. The observed NPR for all four cases with preselector in was in the range 71 - 72 dB. 

For all cases, the observed NPR with the band-limiting filter was 4 to 5 dB higher than without. This may indicate passive IMD in the preselector filters. 

I concur that any comparison with the theoretical maximum NPR value is relevant only to the specific case where the applied noise bandwidth equals fs/2.

Cheers for now, 73,
Adam VA7OJ/AB4OJ

-----Original Message-----
From: Alex VE3NEA [mailto:alshovk at dxatlas.com] 
Sent: 6-Jun-12 13:58
To: Adam Farson; 'HPSDR Reflector'
Subject: Re: [hpsdr] NPR of Mercury (was: Boat Anchors..PreAmps)

Adam,

This sounds like a great plan! Please keep me informed of your findings.

Personally I like the alternate method of NPR estimation described in your paper, where you compare the noise level in the idle channel and outside of it on the band scope directly, without all intermediate steps. Of course, the noise loading must be scaled by the noise/receiver bandwidth ratio as in the other method.

73 Alex VE3NEA

-----Original Message-----
From: Adam Farson
Sent: Wednesday, June 06, 2012 4:27 PM
To: 'Alex VE3NEA' ; 'HPSDR Reflector'
Subject: RE: [hpsdr] NPR of Mercury (was: Boat Anchors..PreAmps)

Hi Alex,

Many thanks for your comments. I shall repeat my NPR measurements on the Perseus, using different noise bandwidths. I am using Telco surplus Wandel & Goltermann noise generators fitted with 1940, 3886, 5340 and 7600 kHz notch filters and their companion band-limiting filters (2048, 4100, 5600 and 8160 kHz, respectively). I have two of these instruments, one with a 12.5 MHz total noise-band and the other with 25 MHz. These values are with no band-limiting filter inserted. What I can do is test at 25 MHz, then insert band-limiting filters and apply the corresponding correction factors.

I can see that the ideal NPR test setup for a direct-sampling receiver is one in which the applied noise-band cuts off at fs/2. Still, though, the receiver's preselector will limit the bandwidth of the applied noise, so to measure the ADC's NPR the preselector should always be out, or a correction factor of 10*
log10 ({fs/2}/preselector BW) should be applied. To determine whether passive IMD in the preselector is a significant factor, both configurations should be tested and the results compared.

Cheers for now, 73,
Adam VA7OJ/AB4OJ

-----Original Message-----
From: Alex VE3NEA [mailto:alshovk at dxatlas.com]
Sent: 6-Jun-12 12:59
To: Adam Farson; HPSDR Reflector
Subject: Re: [hpsdr] NPR of Mercury (was: Boat Anchors..PreAmps)

Hi Adam,

When NPR is measured in an analog receiver, the extra noise in the idle channel is caused mostly by the combinational products that that are created because the amplitude characteristic of the radio is not a perfectly straight line. The magnitude of such noise is a function of the spectral density of the loading noise. Because of this, the NPR for the analog receivers is defined, quoting your article, as "the ratio of the noise power in the notched band to the power in an equal bandwidth adjacent to the notch".

In the RF sampling receivers, the extra noise is dominated by the clipping noise which depends on the total power of the input signal (or noise loading, in case of the NPR measurements), not by its power density. It is the use of the total power vs. power density that causes confusion.

What happens if you repeat your measurements but change the loading noise bandwidth from 5.6 MHz to, say, 2.8 MHz? for an analog receiver, you will get essentially the same results. For an RF sampling receiver, the results will "improve" by 3 dB, because you are measuring the spectral density of the loading noise, and at the same noise power that is required to produce clipping, the spectral density has doubled. To get rid of this fake improvement, you will have to subtract 3 dB from your measurement. Similarly, when you are testing a receiver with a bandwidth of 61 MHz using the noise with a bandwidth of 5.6 MHz, you should subtract 10 * Log10(61/5.6) = 10.37 dB from your raw result, this will immediately turn the unrealistic NPR of 74 dB into something more plausible.

Does this all make sense?

73 Alex VE3NEA

P.S. I used Graph (http://www.padowan.dk/graph/) to plot my charts.

-----Original Message-----
From: Adam Farson
Sent: Wednesday, June 06, 2012 3:07 PM
To: HPSDR Reflector
Subject: Re: [hpsdr] NPR of Mercury (was: Boat Anchors..PreAmps)

***** High Performance Software Defined Radio Discussion List *****

Larry, Alex,

Many thanks for bring me into the group. I joined because I would like to follow this thread, and also out of a general interest in direct-sampling SDR technology.

The criterion I use for determining the optimum noise loading point in a conventional receiver is that proposed by I2VGO. In which the optimum point is that where the noise level at the audio output rises by 3 dB relative to the noise floor. I found that when testing the Perseus and the HPSDR, I could not use this criterion because the ADC clipped before the audio output increased by
3 dB (as noted in my NPR article).

http://www.ab4oj.com/test/docs/test_npr.pdf  Slide 28

Thus, for the SDR tests I used another criterion proposed by Gianfranco Verbana I2VGO,  i.e. a noise loading level at which the ADC just did not clip over a 10 sec. interval:

http://www.ab4oj.com/test/docs/test_npr.pdf  (I2VGO presentation, Slide 36)

and read NPR directly off the SDR s spectral display:

http://www.ab4oj.com/test/docs/test_npr.pdf  (Slide 33)

My results for the Perseus were very close to those presented in Slide 35 of I2VGO' presentation.

http://www.ab4oj.com/sdr/perseus/perseus_notes.pdf (p. 12)

In the ADI MT-005 paper (referenced below), Eq. 16 gives a correction factor which must be applied if the bandwidth of the noise loading is less than fs/2. 
For the Perseus, where fs = 80 MHz, this correction factor was +8.6 dB in my test setup with a 5.6 MHz noise band limiting filter. Now if I take Alex's calculated NPR value of 65.5 dB and add 8.6 dB, I arrive at 74.1 dB, which is the theoretical value I quoted and is also quite close to my best-case measured value of 75 dB.

Looking forward to your further comments.

Cheers for now, 73,
Adam VA7OJ/AB4OJ

---------- Forwarded message ----------
From: Alex VE3NEA <alshovk at dxatlas.com>
Date: 6 June 2012 08:08
Subject: [hpsdr] NPR of Mercury (was: Boat Anchors..PreAmps)
To: hpsdr at lists.openhpsdr.org

***** High Performance Software Defined Radio Discussion List *****

The maximum NPR of 85.4 dB quoted by Adam was computed [1] for an ideal 16-bit ADC whose noise floor of -98.09 dBFS [2] is determined solely by its quantization noise. Perhaps ideal ADC's were out of stock when Mercury was prototyped ;-) because the radio uses LTC2208, a real 16-bit ADC with a noise floor of only -78 dBFS [3]. This makes a huge difference as far as the NPR is concerned.

I have re-computed the maximum NPR using the same equations as in [1], but taking into account the actual noise floor of the ADC. The new chart [4] has the same curves as in Figure 2 of [1], with two more curves added for the real ADC's. The theoretical maximum NPR that can be achieved with LTC2208 is
66.4 dB.

For comparison, LTC2206-14, a 14-bit ADC used in Perseus, has a noise floor of -77.3 dBFS [5]. Its maximum NPR is 65.5 dB.

The values measured by Adam do not indicate any excessive IMD in the preamp of Mercury. To the contrary, for some reason they exceed the theoretical maximum for a real ADC.

References:
[1] www.analog.com/static/imported-files/tutorials/MT-005.pdf
[2] http://en.wikipedia.org/wiki/DBFS
[3] http://www.linear.com/product/LTC2208
[4] http://dxatlas.com/misc/npr.png
[5] http://www.linear.com/product/LTC2206-14

73 Alex VE3NEA

-----Original Message----- From: Larry Gadallah
Sent: Tuesday, June 05, 2012 3:15 PM
To: lester at veenstras.com
Cc: hpsdr at lists.openhpsdr.org ; lstoskopf at cox.net
Subject: Re: [hpsdr] Boat Anchors..PreAmps

***** High Performance Software Defined Radio Discussion List *****

I was reading Adam, VA7OJ's excellent reports on his NPR testing of various receivers including Mercury (see
http://www.ab4oj.com/sdr/hpsdr/hpsdr_npr_va7grr.pdf) and I was intrigued by this
comment:

"The theoretical maximum NPR for a 16-bit ADC is 85.4 dB, as compared to
74.01 dB for a 14-bit ADC.
Thus, the HPSDR receiver exhibits a significant deviation from the theoretical maximum value (much more so than the Perseus, whose highest NPR value in my August 2011 test was 75 dB at 5340 kHz.)

The RF preamplifier in the Mercury receiver is in the RF signal path at all times.  A 20 dB pad is switched in at the preamp input for the  Preamp out function. Apparently, the noise loading provokes sufficient IMD in the preamp to degrade the NPR as much as 12 dB below the theoretical maximum value. (The optimum noise loading value decreases by 19 dB with the attenuator out, which reflects the inserted 20 dB attenuation pretty closely. )"

I'm not sure I grok the meaning of "noise loading" in that comment, but I'm assuming that he is implying that the attenuator contributes enough noise to trigger some nonlinear operation of the preamp? This leads to the question:
Is the current design of leaving the preamp in the circuit at all times and switching in/out an attenuator preventing Mercury from approaching the theoretical NPR capabilities of the LTC2208? Would it be better to completely switch the preamp in/out of circuit as needed?

Cheers
_______________________________________________
HPSDR Discussion List
To post msg: hpsdr at openhpsdr.org
Subscription help:
http://lists.openhpsdr.org/listinfo.cgi/hpsdr-openhpsdr.org
HPSDR web page: http://openhpsdr.org
Archives: http://lists.openhpsdr.org/pipermail/hpsdr-openhpsdr.org/

--
Larry Gadallah, VE6VQ/W7                          lgadallah AT gmail DOT com
PGP Sig: B5F9 C4A8 8517 82AC 16B6  02B6 0645 69F0 1F29 A512

_______________________________________________
HPSDR Discussion List
To post msg: hpsdr at openhpsdr.org
Subscription help: http://lists.openhpsdr.org/listinfo.cgi/hpsdr-openhpsdr.org
HPSDR web page: http://openhpsdr.org
Archives: http://lists.openhpsdr.org/pipermail/hpsdr-openhpsdr.org/

 1339139499.0