Tests on SDR hardware.
(Jan 24 2018)

Soundcards are designed for music!!

We use them in direct conversion SDRs and find often surprisingly good performance. Results reported by different operators may seem inconsistent sometimes.

Mainstream development is towards direct sampling in the VHF range and many regard units like the Softrock Ensemble more like a toy than a real radio. Major improvements are possible however. Alex, HB9DRI, is working on a 4 channel ADC for usage with direct conversion SDR. The noise floor should be something like 15 dB lower than the best soundcard in the table below when used with a Softrock having the Si570 replaced by a good crystal oscillator.

The table is based on measurements that I have documented in video clips on Youtube. Several cards suffer from problems that probably would not cause any audible problem in music recording but that degrade SDR performance severely. Those cards are at the bottom of the list, they suffer from reciprocal mixing. (See videos.)


Manufacturer    Model             Price USD Bits Speed Sat.   Noise   MDS    NF Video Sat to  Reciprocal mixing Modified SR@8dBm  Reciprocal mixing Modified SR@0dBm
                                   Guessed       (kHz) (dBm)  1 kHz  500Hz              MDS        1kHz  Dynrange effective NF       1kHz  Dynrange    effective NF
                                                              (dBm)  (dBm)  (dB)       (dB)        (dBm)  (dB)       (dB)            (dBm)   (dB)       (dB@20kHz) 
Linkrf          UADC4                     ?  24    96   -18  -129.6  -132.6 14.4 12    114.6      -124.5  119.5      19.5           -118.2   121.2         25.8
Terratec        DMX 6fire USB           350  24    96   -18  -129.7  -132.7 14.3 4,6   114.7      -123.2  118.2      20.8
M-Audio         Delta 44 modified       150  24    96   -18  -127.8  -130.8 16.2  8    112.8      -121.4  116.4      22.6
Creative        X-Fi Surr.5.1 USB mod.   65  24    96   -18  -128    -131   23   11    113        -121.5  115.5       22
Creative        Audigy SE Linux          30  24    96   -18  -127.6  -130.6 16.4 8,9   112.6      -120    115         24
Creative        Audigy SE Windows        30  24    96   -18  -126.1  -129.1 17.9  8    111.1      -120    115         24
Lynx            Lynx Two (mod)          800  24    96   -18  -130.2  -133.2 13.8 3,6   115.2      -120    115         24
Terratec        DMX 6fire PCI           100  24    96   -18  -128.3  -131.3 15.7 3,7   113.3      -119.5  114.5      24.5
Realtek         X9DAI motherboard            24    96   -18  -126.9  -129.9 17.1  8    111.9      -119.3  114.3      24.7
Xonar           Essence STX             150  24    96   -18  -124    -127    20  2,10  109        -118.7  113.7      25.3
M-Audio         Revolution 5.1          100  24    96   -18  -127.8  -130.8 16.2 3,7   112.8      -118.3  113.3      25.7
AKAI            EIE pro                 250  24    96   -18  -126.2  -129.2 17.8 4,7   111.2      -118    113         26
M-Audio         M-Track Quad            270  24    96   -18  -126.2  -129.2 17.8 5,6   111.2      -117    112         27
Creative        SB Live!                 20  16    48   -18  -121.8  -124.8 22.2 4,10  106.8      -114.5  109.5      29.5
M-Audio         Delta 44                150  24    96   -18  -124.3  -127.3 19.7 1,7   109.3      -114.1  109.1      29.9
IDT (Sigmatel)  D5400XS mtherbd              24    96   -21  -126.8  -129.8 17.2 1,8   108.8      -117.1  109.1      26.9
IDT (Sigmatel)  D5400XS mtherbd-3dB          24    96   -18  -125.2  -128.2 18.8 1,8   110.2      -114    109         30
ESI             Maya 44 Xte             180  24    96   -18  -124.4  -127.4 19.6 2,7   109.4      -112.9  107.9      31.1
Steinberg       UR12                    120  24    96   -20  -127.9  -130.9 16.1 5,7   110.9      -114    107         30
Creative        SB Live! 5.1 Digital     20  16    48   -18  -117.2  -120.2 26.8 4,10  102.2      -109.4  104.4      34.6
SIIG            CE-SA0011                30  24    96   -21  -118.9  -121.9 25.1 1,8   100.9      -111    103         33
Creative        SB Live! External USB    80  24    96   -18  -124.9  -127.9 19.1 5,10  109.9      -105.7  100.7      38.3
StarTech        Virtual 7.1 USB          27  24    96   -20  -121.5  -124.5 22.5 1,7   104.5      -106    99          38
Komplete        Audio 6                 250  24    96   -18  -128.1  -131.1 15.9 4,6   113.1       -80    75          64
Syntek          Easier CAP                7 	    8                             1

Reciprocal mixing is measured in 1 kHz bandwidth on a modified Softrock with a 10 dB stronger signal. Measurement is done at 20 kHz separation on cards with normal behaviour but at the frequency of maximum noise on cards with abnormal behaviour. See the respective video for details.

Videos: [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12]

What does the table mean?

Antennas receive noise of several kinds. The lowest noise power we can expect below 15 MHz is about 27 dB above the power of a room temperature resistor. This means that with a noise figure (NF) of 27 dB we might have 50% of the noise from the antenna and 50% from our receiver. That means a loss compared to a noiseless receiver would be 3 dB. That would be in quiet rural locations only. Look at figure 2 in this ITU paper: RECOMMENDATION ITU-R P.372-13

Allowing the noise floor to rise by more than 7 dB (5 times) is a waste of resources. The antenna temperature, Tant would never be below 500 times the room temperature. The lowest meaning ful Rx noise temperature Trx would then come from this relation:

Tant + Trx = 5 Trx

From this we conclude that a Rx noise temperature below 125 times the room temperature is not useful. This means that we do not want a noise figure below 21 dB. From the table we can immediately see that most soundcards have a low enough noise floor when used with a Softrock Ensemble that has not been modified at all. If one uses the Softrock Ensemble as is there is no reason at all to buy any good soundcard. One just has to avoid the really bad ones.

The "working dynamic range" is the distance from the noise floor, 27 dB above room temperature or -120 dBm in a bandwidth of 500 Hz to the point of saturation. With the original Softrock Ensemble, most soundcards give saturation at -18 dBm. That means that the working dynamic range is 102 dB with the usual bandwidth 500 Hz for MDS (minimum discernible signal = the noise floor.)

We thus need MDS (of the card itself) to be at least 108 dB below saturation for systems that saturate at -18 dBm. The column "Sat to MDS" shows that almost all soundcards meet this requirement. There is however one more thing. Some cards suffer from reciprocal mixing. The column Dynrange under "Reciprocal mixing. Modified SR" must also show at least 108 dB. A couple of soundcards fail really badly:
Creative SB Live! External USB 100.7 dB
StarTech Virtual 7.1 USB 99 dB
Komplete Audio 6 75 dB (!!!!)

Some are not what one would like to buy for SDR:
ESI Maya 44 Xte 107.9 dB
Steinberg UR12 107 dB
Creative SB Live! 5.1 Digital 104.4 dB
SIIG CE-SA0011 103 dB

Simple modifications.

If one buys a Terratec DMX 6fire USB one gets a NF of 14.3 dB with the un-modified Softrock Ensemble. To make it fit a low noise antenna one should place 7 dB of attenuation in front of the Softrock Ensemble. That would move the point of saturation to -11 dBm with a working dynamic range of 109 dB if reciprocal mixing were ignored. To really get this improvement it would be necessary to add capacitors near the Si570 chip to remove USB noise as demonstrated here: rxdynrange_softrock

My modified Softrock Ensemble with 10 dB less gain shows a dynamic range of 118.2 dB with reciprocal mixing included. The NF is 20.8 dB, just what we want, and the working dynamic range is 112 dB. To really get that performance one might often need a preselector because the mixer would saturate if strong wideband signals are allowed into it. In situations where the strong signals are within the amateur band of interest, there would be no need for preselectors.

The Creative Audigy SE is about 10 times cheaper but nearly as good. It may be difficult to use because of driver problems, look at the videos for details.

Not so simple modifications.

The WSE RX2500 was designed in year 2001 with components of that era. The mixer uses 74HC4052 with about 70 ohm switch resistance as compared to 4 ohms in the FST3253 used in the Softrock Ensemble. Yet, mixer performance is dramatically better as you can see here: uadc4wse Noise is 8 dB lower and harmonics -70dB as compared to -45 in my most modified Softrock. What I see is reciprocal mixing noise floor@10kHz = -156 dBc/Hz. That is something like 10 dB better than the Perseus. It uses the UADC4 much better than the Softrock Ensemble, but improvements are still possible. The very sharp anti-alias filtering provides some noise and a too high gain. By running the UADC4 at 192 kHz one could move the anti-aliasing into the digital domain and get a significantly better performance. A low Q anti-alias cutting at 96 kHz would not have to add gain or noise. Further, the UADC4 provides some filtering above 96 kHz.

Change 1: Connect the mixer output to the plus input of the OP-amp and use a low impedance voltage divider from the output to the minus input to set gain. Would make the current through the switch much lower and thereby reduce distortion/overtones. Change 2: (Alternatively) Put an inductor in parallel with R58 which could be made 100 ohms or so. Put a capacitor from the minus input of the OP-amp to ground. Possibly remove C42.move feedback R59, C45 to the midpoint of the secondary of T5 and put the gain-determining resistor between C70 and the well decoupled Vcc/2. One more secondary winding would be needed for T5. Compare with the RX2500 mixer: The AF feedback four phase CMOS switch mixer There are two transformers, TR1 and TR2. They provide four phases of the RF signal. Remove one of them and also remove the second OP-amps, IC4 and IC11 and you have the schematic I propose for the Softrock with a single transformer having two secondary windings.

Conclusions.

Popular trends today say direct sampling is SDR for the future. The UADC4 demonstrates that direct conversion today can be better. The direct conversion architecture of Softrock would need a lower noise LO, a better mixer and good front end filtering to become better than the best direct sampling radios. They also need good RF filtering...

The direct conversion radios can use polyphase mixers. (The WSE uses 4 phases, but much more is possible.) The first example is the Airspy HF+ A very interesting radio provided a front end filter is used to remove signals that could cause second order intermodulation. (something most radios need, but not Perseus because such filters are alredy included inside the box.) A firmware change might make the HF+ performance good enough even without filters with respect to the findings here: Peter E Chadwick, G3RZP, HF Receiver Dynamic Range: How Much Do We Need? QEX May/June 2002, p 36-41. The future of SDR hardware is interesting....