Linrad support: High performance hardware for Linrad.
(July 13 2006)

Many boxes

The dedicated hardware for Linrad which is presented here is high performance hardware for the experimenter. It takes the form of several well defined functional blocks with 50 ohm impedance for input and output signals. This way it is possible to replace single blocks at a later date when technology has improved. Well screened units for different functions makes it easier to avoid spurious responses. No attempt has been made to minimise the box sizes or power consumption. It is very probable that the same performance can be obtained with smaller boxes at lower power consumption and lower cost. My time is a limitation and I happily adopt the first solution I can find that is good enough. Below is a short description of the different units and links to the full descriptions.

Some day I will (hopefully) make transmit hardware by use of the same strategy. Tx outputs will fit the rx inputs in frequency and amplitude so rx and tx performance can be verified easily at any point in the rx/tx signal chain.

Buying components and PCB's in small nubers is expensive. Building a single set of converters would become very expensive if components have to be bought in very small quantities on the open market. In a joint effort between me (SM5BSZ) and Niklas (SM7UFW) we make the units presented here available under the trademark WSE (Weak Signal Equipment).

The A/D board

In the development phase I have been using a Delta 44 soundcard to sample four analog channels at 96 kHz. The Delta 44 is a so called 24 bit soundcard which does not mean that that it gives 24 bits of valid data. The Delta 44 gives 16 bits of correct data and two or three more bits are needed to give the analog noise floor of the A/D converter itself with an accuracy high enough to not add any digitalization noise. The remaining 5 or 6 bits are superflous.

The Delta 44 can be modified to reduce the noise floor between 2 and 4 dB. The noise floor varies between different cards but they are all practically identical when modified. The modified Delta 44 provides a dynamic range of 148 dBc/Hz with harmonic spurs that are about 85dB below the carrier.

The Lynx Two soundcard is better than the Delta 44 by about 10 dB but it is only 7 dB better than the modified Delta 44. When used together with a WSE RX2500, the improvement obtainable with the Lynx Two card is only 4 dB becaust the op-amps in RX2500 are fed with +/-12V only so they can not drive the Lynx Two higher than 3 dB from saturation. Lynx Two supports 192 kHz sampling and the larger bandwidth is very interesting. An investigation about how to modify the RX2500 for 192 kHz bandwidth and how to improve the Lynx Two soundcard shows that the analog input of the Lynx Two can easily be improved by 7 dB by removal of the needless input stages (The RX2500 contains the differential amplifier needed to reference the analog signal to the proper ground point on the Lynx Two.) For details, look here: Modifying RX2500 and Lynx Two for optimum SDR system performance

As it turns out, the sampling clock of the Lynx two is totally useless for SDR purposes. The phase noise is extremely high, when a signal is placed near the upper end of the passband the phase noise of the sampling clock gives rise to a wideband noise floor at -135 dBc/Hz. This is no better than a cheap 16 bit soundcard. The A/D converter of the Lynx Two, the AK5394 is EXTREMELY good however and it is used in many high end soundcards. Maybe there is one that uses a crystal oscillator for the sampling clock?

Hardware for fixed passband 2.455MHz to 2.545MHz

RX2500, the 2.5MHz to audio converter is designed to fit the Delta 44 soundcard It nearly preserves the dynamic range of the modified Delta 44, it only adds about 1dB to the noise floor.

For signals outside the visible passband the dynamic range is nearly 20dB better.

Hardware for 25kHz interleaved passbands at 10.7MHz

RX10700, the 10.7 to 2.5MHz converter is designed to fit RX2500. It uses 4 crystals with 25kHz frequency separation as the local oscillator. This way oscillator phase noise is avoided. These four frequency bands can be selected.

10.630 to 10.720
10.655 to 10.745
10.680 to 10.770
10.705 to 10.795

This unit adds 1dB to the noise floor but it does not degrade the intermodulation characteristics. At very high signal levels, about 160dB above the noise in 1Hz bandwidth, this unit produces higher order spurs that arise from overtones of 10.7MHz that are mixed with overtones of the local oscillator. At a first glance it may look like overkill to have these spurs as far down as at -160dBc/Hz but in a real weak signal usage the the preamplifier will add 20dB to the noise floor which means that these spurs will show up somewhere around -142dBc/Hz.

Hardware for 25kHz interleaved passbands at 70MHz

RX70, the 70 to 10.7 MHz converter is designed to fit RX10700. With five crystals separated by 100kHz one gets 20 interleaved bands with 25kHz separation. The following frequency bands can be selected with a receiver built from the three units RX70, RX10700 and RX2500 connected to a 96kHz soundcard.

69.930 to 70.020 (0,0)
69.955 to 70.045 (1,0)
69.980 to 70.070 (2,0)
70.005 to 70.095 (3,0)
70.030 to 70.120 (0,1)
...
...
...
70.380 to 70.470 (2,4)
70.405 to 70.495 (3,4)

Since only crystal oscillators are used, the phase noise of the local oscillators is very low, reciprocal mixing causes a noise level of -171dBc/Hz for the complete 70MHz receiver at frequency separations of 100kHz and -175dBc/Hz at 500kHz. Within the 90kHz passband, the noise floor is 146 dB/Hz below the saturation limit of the Delta44.

Ham band converters

It is fairly straightforward to build high dynamic range converters for the different amateur bands using one crystal for each band.

If the converter to 70MHz is allowed to degrade the noise floor by 2dB the system will have its noise floor somewhere around 143dBc/Hz within the 90 kHz passband. This number is similar to what one can expect from good analog receivers. For larger frequency separations, the dynamic range should be much better for the hardware described here.

When the Linrad system is used for weak signal traffic, the noise floor is degraded by 20dB because the noise from the mast mounted preamplifier has to be the origin of 99% of the noise if the last 0.1dB of possible S/N is desired. The 123dBc/Hz that then remains corresponds to a dynamic range of 90dB in SSB bandwidth which is adequate under most circumstances. 10dB less RF gain will improve the dynamic range by 10dB while the S/N is degraded by about 0.4dB.

Hardware for 25kHz interleaved passbands at 144MHz

RX144, the 144 to 70 MHz converter is designed to fit RX70. With four crystals separated by 500kHz one gets 80 interleaved bands with 25kHz separation. The following frequency bands can be selected with a receiver built from the four units RX144, RX70, RX10700 and RX2500 connected to a 96kHz soundcard.

143.930 to 144.020 (0,0,0)
143.955 to 144.045 (1,0,0)
143.980 to 144.070 (2,0,0)
144.005 to 144.095 (3,0,0)
144.030 to 144.120 (0,1,0)
...
...
...
145.880 to 145.970 (2,4,3)
145.905 to 145.995 (3,4,3)

Hardware for the lower HF bands

RXHFA, the lower ham bands to 70 MHz converter is designed to fit RX70. With five crystals, one for each amateur band 1.8 MHz, 3.5 MHz, 7 MHz, 10 MHz and 14 MHz, one gets 20 interleaved bands with 25kHz separation that give coverage over 565 kHz for each band. The following frequency bands can be selected on 3.5 MHz with a receiver built from the four units RXHFA, RX70, RX10700 and RX2500 connected to a 96kHz soundcard.

3.430 to 3.520 (0,0,1)
3.455 to 3.545 (1,0,1)
3.480 to 3.570 (2,0,1)
3.505 to 3.595 (3,0,1)
3.530 to 3.620 (0,1,1)
...
...
...
3.880 to 3.970 (2,4,1)
3.905 to 3.995 (3,4,1)

The coverage is from (X - 70) kHz to (X + 495) kHz where X is one of the five amateur bands that the unit covers.

Frequency and gain control

The WSE units use a simple serial interface to set frequency and for the RXHFA unit also gain. For details look here: Controlling WSE units from within Linrad

Dynamic range

Receiver dynamic range may be limited by several different factors.

One limitation is the linearity of amplifiers, mixers, filters and everything else in the signal path. The related performance limitation is often expressed as intercept points, IP2 and IP3 for second and third order intermodulation respectively.

Another limitation is oscillator purity. Any noise sideband associated with local oscillators will produce noise sidebands around strong signals. A phenomenon known as reciprocal mixing.

Yet another limitation is amplitude or phase modulation added by amplifiers. RF amplifiers must have well decoupled supply voltages and they must be designed to carry very low noise current at audio frequencies to not add noise sidebands. Modulation added by amplifiers degrade BDR and is difficult to distinguish from reciprocal mixing.

A good receiver needs a dynamic range of about 100dB according to a study by Peter E. Chadwick, G3RZP HF Receiver dynamic Range: How Much Do We Need? which appeared in the May/June 2002 issue of QEX.

The number 100dB for phase noise limited dynamic range given by G3RZP refers to "SSB bandwidth" which means that the noise floor of a good receiver has to be below -132dBc/Hz (3dB S/N loss, 3kHz bandwidth). The phase noise limited dynamic range of the Linrad hardware at the 10.7MHz IF input is -145dBc/Hz or -148dBc/Hz depending on whether one or two channels contain the offending signal. The limit is mainly determined by the dynamic range of the modified Delta 44 A/D converter and not by the Linrad hardware. In case the Delta 44 soundcard is not modified the dynamic range is 2 to 4 dB smaller and some broad spurs are present that degrade by 10dB or more at certain frequencies.

The number 96dB for intermodulation limited dynamic range is strongly affected by receiver arcitecture, how early in the signal path selectivity is introduced. IP3 values will improve as the separation of the offending signals from the desired signal increases.

The intermodulation limited dynamic range can not be applied directly to an SDR (software defined radio) as Linrad. As long as the A/D converter is not saturated, very low levels of intermodulation are produced but if saturation occurs, the interference level abruptly becomes extremely high. The Linrad hardware has high performance filters to protect the A/D converter from signals outside the desired 95kHz passband. A comparison with conventional receivers can not be done using a single number.