Comparison of “conventional”, “DDC based” and “soundcard based” receivers

Back to main page

(Below is a cleaned up version of a rambling message I fired off to the Softrock40 and the Mobokits discussion groups on 22 July 2010)

Conventional Superhets, DDC based receivers and I/Q demod + soundcard based receivers all have one thing in common; they are used to separate an -often very weak- RF signal from the sum total of all signals present at the antenna input. A basic property which makes or breaks any receiver is distortion-free dynamic range and selectivity. On HF, sensitivity is usually not much of an issue, unless the design is wrong, then the system needs more gain.

Dynamic range:

When it comes to dynamic range, a classic analog superheterodyne receiver and a soundcard based receiver have one thing in common, the wide open front end and the first mixer. This places the burden on front end and mixer to have a good dynamic range, any nonlinearities caused by strong signals somewhere in the passband will cause any signals to mix with any other signals and cause a multitude of phantom crud. This is why IMD or IP3 measurements are such a good measure of receiver quality. For DDC receivers, see discussion further down.

In a superhet receiver, you will see a tuned or untuned RF-stage with fixed gain, normally followed by a mixer to the IF. Usually you have then a crystal filter and automatic gain control (AGC) circuit immediately behind the first mixer, making any dynamic range issues far less strenuous. And of course a modern superhet usually ends with a DDC demodulator (often, in my opinion, misnamed as a DSP in this context as it is implemented with a digital signal processor), rather than the classic narrow band crystal filter arrangement for CW and SSB reception. So called roofing filters in front of the DDC are often there, at least in part, to overcome the limited bit resolution of the DDC, by limiting the bandwidth it has to work with.

In a soundcard based receiver however, the limitation of dynamic range is caused almost entirely by the soundcard, there is no interactive AGC following the mixer.

16 bit soundcards:

A 16 bit soundcard has a maximum theoretical dynamic range of 6.02 x 16 - 3 = 93.3 dB when injected with a pure sine wave. However, it is more realistic to look at the sum aggregate power of all signals present within the passband of the soundcard. This can be explained as follows:

With A/D sampling, one needs to take care that the sum of signals present does not exceed the maximum input value. If it does, then the sampled digital representation of all signals present will be hopelessly distorted.

Inject one pure sine wave and you get the full dynamic range, add another equal strength pure sine wave at a different frequency within the passband, and in order not to overload the A/D, the overall dynamic range for each sinewave is now 6 dB less. Inject a busy spectrum of signals within the passband of 48/96/192 kHz, and the channel you are interested in will have that much less dynamic range to work with.

In a 48 kHz bandwidth, one would get a an absolute worst case scenario by comparing the proportion of the overall bandwidth with the actual receive window bandwith, say 3 kHz.

Integrating across the whole bandwidth, you get 10log48000 = 46.8 dB.

Integrating across 3 kHz you get 10log3000 = 34.8 dB.

In other words, the dynamic range becomes:

(93.3-(46.8 -34.8)) = 81.3 dB,

which translates to 13.5 S units on an S-meter (S0 – S9+27dB). This of course is correspondingly worse by 3 or 6 dB when using 96 or 192 kHz bandwidths.

The above example postulates a 16 bit card at its maximum capabilities. In reality the dynamic range of the sound card will be worse than 96 dB, probably worse by 10 dB.

Now we arrive at a total dynamic range of 81.3 dB – 10 dB = 71.3 dB, not great.

In addition, unless the soundcard has some serious input filtering, including digital oversampling, then in reality there will be aliasing products as a result of sampling from a far greater bandwidth, the overall results being not so impressive.

Note: A separate issue is that of measured signal or noise power within a defined sub-bandwidth of the total sampled digital signal, measured in dBfc/Hz. The theoretical minimum noise power measured in a channel in dBfc/Hz will be a 1Hz portion of the minimum theoretical noise present in the whole bandwidth (for a 16bit card at 48kHz):

-6.02 dB x 16 bits - 3 dB - 46.8 dB = -141.92 dBc. A thoroughly misleading number :)

24 bit sound cards:

Enter the good quality 24 bit soundcard with multiple oversampling:

24 bit soundcards come in many flavours and qualities, none approaching the theoretical maximum. I have tested some mediocre ones and some pretty darn impressive ones.

To use the same analysis as before, max dynamic range: 6.02 dB x 24 bits - 3 dB = 141.48 dB, lets shave 20 dB off this number for worst case realism: 121 dB.

Worst-worst case soup of signals present: 121 – (46.8 – 34.8)) = 109 dB, not bad at all, corresponds to 18 S-units or a range of S0 to S9 + 55dB on our trusty S-meter. (48 kHz bandwidth, subtract additional 3 or 6 dB for 96 or 192 kHz)

The above stipulates of course that your receiver gain has been adjusted to fit the soundcard capabilities. Add an input attenuator like those on commercial HAM radio superhets, and you are covered for all situations. I believe we now have a similar performer to those superhets, at least in terms of dynamic range.

Below is a description of a simple test to determine the dynamic range of a Softrock 6.3/Mobo/24 bit soundcard setup.

The soundcard used for this test was the SDR-Widget Lite (Alpha2).

In order to take advantage of the dynamic range of this sound card, the audio amplifier loop gain in the Softrock 6.3 receiver was reduced by 25dB (R53 and 54 at 274ohm) .

The PC was running Linux (kernel 2.6.35-rc3) and SDR software was Linrad by SM5BSZ (see also writeup on DDC receivers)

A good quality ovenized 10MHz OCXO was used to inject a signal into the SR6.3/Mobo receiver.

With this setup, the RF signal input was adjusted up to +4dBm, which was found to be the maximum level possible before the audio input to the audio card would go into distortion (break hopelessly apart). This would give just over 120dB Signal over the noise floor and very little noticeable increase in the noise floor itself (5dB). See the screenshot below which shows the signal in a +/-3kHz bandwidth.

One of the first things apparent from this screenshot, and perhaps the largest surprise for me, was the superb phase noise performanc of the Si570 oscillator. The screenshot shows the noise floor at 3kHz out to be already down to approximately -118dBm.

With this test, an input signal at +4dBm, some compression in the QSD circuit was noted.

When testing with a slightly increased audio amplifier gain in the SR 6.3, with R53/R54 set at 499ohm (20dB decrease), the compression was found to be less at full level (now reduced to approx -1dBm) and the noise floor between 500Hz and 3kHz out to the side, was found to be almost flat. See here

Below is another screen shot, this time also showing a second signal, 2kHz up, or at 10.002MHz. The second signal is 105dB weaker, or at -101dBm (2uV/50ohm). As can be seen, this signal is easily visible, a feat that is probably not repeatable by any superhet receivers within normal budget.

As can be seen on this screenshot also, the -101dBm signal at 2kHz up is easily visible at about +15dB over the noise floor. However due to intermodulation in the QSD, caused by the very strong signal at 10.000MHz, an image at a weaker strength is also visible at 2kHz down.

Hey, we are up to a dynamic range of S9 +74dB

I/Q demodulator weaknesses:

One of the weaknesses of the Soundcard based SDR is the I/Q modulation/demodulation scheme which requires a near constant 90 degree phase difference across a large spectrum of input frequencies. Any errors will affect the image suppression and in the worst case, an almost equal strength mirror image will be produced.

Errors can be compensated for in the SDR software, but this can be cumbersome if the errors vary much across the HF band as well as across the audio input band. Also, any inequalities in the analog path before the A/D, especially in cheaper audio cards, may introduce inequal signal amplitude or inequal group delays (frequency dependent phase errors) between the two input channels, not to mention phase errors and jitter that may be introduced as well. A good soundcard will however in practice overcome all of those errors, except of course the errors introduced in the I/Q demodulator itself, those can be addressed by the SDR software. In practice it appears not difficult to achieve image suppression of 60 dB or more.

Lack of crystal filters in an SDR:

Crystal filters are simply narrow band mechanical filters. These introduce group delays (frequency dependent phase delays), especially if they are very narrow. This phenomenon is experienced as “ringing”.

In an SDR implementation you will normally see an Infinite Impulse Response (IIR) filter or its more computationally intensive brother, Finite Impulse Response (FIR) filter, used. An IIR filter utilizes feedback, a sharp filter can be implemented with not too many poles, similar to an analog filter. This filter also introduces group delays, similar to an analog filter. A FIR filter however, does not implement feedback, does not cause group delays, but requires more iterations and is therefore more resource intensive. With a FIR, a near brick-wall like filter can be implemented, the penalty (there is always one) being signal processing delays.

In conclusion, a filter implemented through DSP techniques is no worse than a crystal filter, and can be made better.

A very brief take on Direct Digital Conversion (DDC) receivers:

DDC receivers simply sample the RF at the antenna output and the same holds true as before, the receiver needs to separate a -often very weak- RF signal from the sum total of -often very large- signals present at the antenna input.

Most of the currently very popular DDC receivers on the market now use 14 bit A/D and some use 16 bit A/D. This limitation can be improved by bandwidth decimation which will increase the S/N by eliminating quantization noise in the bandwidth not carried across. Below is a comment from Roelof Bakker, PA0RDT, showing how this works for a DDC receiver implementing a 14 bit A/D conversion:

Due to to non perfect circuit implementation, a 14 bit A/D converter performs in practice as a 13 bit device at best. This will give a dynamic range of:

20log(2^12.8) = 76 dB. (comment TF3LJ: data sheet for LTC2206 states 78.2dBfs)

The LTC2206 samples at 80 MS/s and clips at -6 dBm, so the noise floor will be -82 dBm.

Due to oversampling the noise floor will improve Gp = 10log(Fs/2xB)

For a SSB bandwidth of 2.4 kHz, Gp = 42 dB and the noise floor will be 82 + 42 = -124 dBm

This equals a noise figure of 16 dB. The dynamic range is 104 dB.

Though the 16 dB noise figure looks rather poor, in practice it is sufficient for reception on 10m with my doublet (2 x 16 m top, 11m long feeders, balanced ATU). The concept of third order intercept point is a bit misleading in regard to A/D converters as these stop abruptly when the overload point has been reached.

In the above example, the LTC2206-14 (a 14 bit A/D device) is being used. The LTC2206 comes in 14 and 16 bit versions, however both versions are similar in performance, at slightly worse than 13 bits (12.8). The HPSDR Mercury project uses the LTC2208, another 16 bit A/D which can sample at up to 130 MS/s. This A/D has a similar noise floor performance to the LTC2206 or -78dBfs, which equals 78/6.02 = ~13 bits effectively. Thus the Mercury will have a similar performance to the one stated in the above example.

Here is a link to performance tests made by SM5BSZ, testing three DDC receivers, the SDR-14, Perseus and SDR-IP (option 01). These results confirm the above example in illustrating the good dynamic range achievable with DDC type receivers. When comparing my own test on a Softrock 6.3/Mobo/SDR-Widget Lite (Alpha) 24 bit Audio card, as described further up on this page, it can be seen that the audio card receiver and the DDC receiver are comparable in this respect, both better than the conventional superhet receiver.

One could say that the DDC receiver is the most elegant SDR solution, RF in, audio out... all the analog stuff gone. In comparison with the audio I/Q method, these receivers are rather expensive though.

Epilog:

Oh... I cannot resist but to mention: The ITU-R definition for SDR is somewhere along the lines of a radio which parameters can be altered through an over-the-air software upload. In other words, we are not working with SDR but rather Software Enabled Radios :))) I am not current on whether this definition has been carved in stone yet, but this is the last I heard.

In conclusion:

I do not agree with some opinions voiced, that the Softrocks or other soundcard based receivers can never compare with superhets :))

This spur of the moment rambling was instigated by a discussion on the Yahoo Groups Softrock40 reflector.

Back to main page

(Page last updated 2010-07-28, Loftur E. Jonasson, TF3LJ)As can be seen on this screenshot also, the -101dBm signal at 2kHz up is easily visible at about +15dB over the noise floor. However due to intermodulation in the QSD, caused by the very strong signal at 10.000MHz, an image at a weaker strength is also visible at 2kHz down.

Hey, we are up to a dynamic range of S9 +74dB

I/Q demodulator weaknesses:

One of the weaknesses of the Soundcard based SDR is the I/Q modulation/demodulation scheme which requires a near constant 90 degree phase difference across a large spectrum of input frequencies. Any errors will affect the image suppression and in the worst case, an almost equal strength mirror image will be produced.

Errors can be compensated for in the SDR software, but this can be cumbersome if the errors vary much across the HF band as well as across the audio input band. Also, any inequalities in the analog path before the A/D, especially in cheaper audio cards, may introduce inequal signal amplitude or inequal group delays (frequency dependent phase errors) between the two input channels, not to mention phase errors and jitter that may be introduced as well. A good soundcard will however in practice overcome all of those errors, except of course the errors introduced in the I/Q demodulator itself, those can be addressed by the SDR software. In practice it appears not difficult to achieve image suppression of 60 dB or more.

Lack of crystal filters in an SDR:

Crystal filters are simply narrow band mechanical filters. These introduce group delays (frequency dependent phase delays), especially if they are very narrow. This phenomenon is experienced as “ringing”.

In an SDR implementation you will normally see an Infinite Impulse Response (IIR) filter or its more computationally intensive brother, Finite Impulse Response (FIR) filter, used. An IIR filter utilizes feedback, a sharp filter can be implemented with not too many poles, similar to an analog filter. This filter also introduces group delays, similar to an analog filter. A FIR filter however, does not implement feedback, does not cause group delays, but requires more iterations and is therefore more resource intensive. With a FIR, a near brick-wall like filter can be implemented, the penalty (there is always one) being signal processing delays.

In conclusion, a filter implemented through DSP techniques is no worse than a crystal filter, and can be made better.

A very brief take on Direct Digital Conversion (DDC) receivers:

DDC receivers simply sample the RF at the antenna output and the same holds true as before, the receiver needs to separate a -often very weak- RF signal from the sum total of -often very large- signals present at the antenna input.

Most of the currently very popular DDC receivers on the market now use 14 bit A/D and some use 16 bit A/D. This limitation can be improved by bandwidth decimation which will increase the S/N by eliminating quantization noise in the bandwidth not carried across. Below is a comment from Roelof Bakker, PA0RDT, showing how this works for a DDC receiver implementing a 14 bit A/D conversion:

Due to to non perfect circuit implementation, a 14 bit A/D converter performs in practice as a 13 bit device at best. This will give a dynamic range of:

20log(2^12.8) = 76 dB. (comment TF3LJ: data sheet for LTC2206 states 78.2dBfs)

The LTC2206 samples at 80 MS/s and clips at -6 dBm, so the noise floor will be -82 dBm.

Due to oversampling the noise floor will improve Gp = 10log(Fs/2xB)

For a SSB bandwidth of 2.4 kHz, Gp = 42 dB and the noise floor will be 82 + 42 = -124 dBm

This equals a noise figure of 16 dB. The dynamic range is 104 dB.

Though the 16 dB noise figure looks rather poor, in practice it is sufficient for reception on 10m with my doublet (2 x 16 m top, 11m long feeders, balanced ATU). The concept of third order intercept point is a bit misleading in regard to A/D converters as these stop abruptly when the overload point has been reached.

In the above example, the LTC2206-14 (a 14 bit A/D device) is being used. The LTC2206 comes in 14 and 16 bit versions, however both versions are similar in performance, at slightly worse than 13 bits (12.8). The HPSDR Mercury project uses the LTC2208, another 16 bit A/D which can sample at up to 130 MS/s. This A/D has a similar noise floor performance to the LTC2206 or -78dBfs, which equals 78/6.02 = ~13 bits effectively. Thus the Mercury will have a similar performance to the one stated in the above example.

Here is a link to performance tests made by SM5BSZ, testing three DDC receivers, the SDR-14, Perseus and SDR-IP (option 01). These results confirm the above example in illustrating the good dynamic range achievable with DDC type receivers. When comparing my own test on a Softrock 6.3/Mobo/SDR-Widget Lite (Alpha) 24 bit Audio card, as described further up on this page, it can be seen that the audio card receiver and the DDC receiver are comparable in this respect, both better than the conventional superhet receiver.

One could say that the DDC receiver is the most elegant SDR solution, RF in, audio out... all the analog stuff gone. In comparison with the audio I/Q method, these receivers are rather expensive though.

Epilog:

Oh... I cannot resist but to mention: The ITU-R definition for SDR is somewhere along the lines of a radio which parameters can be altered through an over-the-air software upload. In other words, we are not working with SDR but rather Software Enabled Radios :))) I am not current on whether this definition has been carved in stone yet, but this is the last I heard.

In conclusion:

I do not agree with some opinions voiced, that the Softrocks or other soundcard based receivers can never compare with superhets :))

This spur of the moment rambling was instigated by a discussion on the Yahoo Groups Softrock40 reflector.

Back to main page

(Page last updated 2010-07-28, Loftur E. Jonasson, TF3LJ)

As can be seen on this screenshot also, the -101dBm signal at 2kHz up is easily visible at about +15dB over the noise floor. However due to intermodulation in the QSD, caused by the very strong signal at 10.000MHz, an image at a weaker strength is also visible at 2kHz down.