linrad support: RF amplifier and filter for 2.5MHz
(April 19 2004)

The noiseless feedback amplifier

By use of transformers one can apply both voltage feedback and current feedback. There was an article in Ham Radio november 1979 by DJ2LR Ulrich Rohde giving details about this circuit.

For the 2.5MHz to audio converter the transformation ratio from drain to source is 2:1. The output transformer, TR3 in fig. 1, is 9 turns trifilar wound on a 16 mm toroid. Two of the windings are connected in series at the drain side. The drain is loaded by one 100 ohm resistor to ground and one 100 ohm resistor to the bandpass filter. For details, see the circuit diagram of fig. 1. The input impedance of the filter is about 100 ohms within the passband but it varies a little. Outside the passband the input impedance of the filter is lower.

The impedance at the drain of the transistor is very low due to the feedback. The impedance feeding the filter is therefore 100 ohms regardless of what impedance is connected to the input of the amplifier.

The third order input intercept point, IP3 of the amplifier is about 40 dBm. The 1dB compression point is at +15 dBm when clipping occurs at the drain. The RF voltage at the drain is then 20V which makes the power absorbed in the 100 ohm resistor 0.4W. Two input signals of +8 dBm give third order IM at -55dBc while two signals at 0 dBm gives IM3 at -81dBc

The input transformer, TR5 in fig. 1. steps up the voltage by a factor of 2.5 from the antenna input to the gate. The source current is routed through a small winding to provide a load on the transformer that makes the input impedance 50 ohms.

The input transformer is loaded by the input capacitance of the transistor, the inductance of the transformer balances for a low Q resonance at 2.5 MHz. By reducing the number of turns on the transformer and adding a capacitor at the input, C116 in fig. 1, it would be possible to make the Q of the input resonance higher to attenuate signals well away from 2.5 MHz. The final design does not use any capacitor at C116 because 2.5 MHz is an IF frequency and the stages in front of it can not emit any signal well away from 2.5 MHz at a level that would be a problem for the 2.5 MHz amplifier. High Q in the input transformer is meaningless in this case.

The input transformer, TR5, is wound on a ferrite toroid core from Ferroxcube (Philips). Material 4C65, type TN 14/9/5. The gate winding is 31 turns, the source winding is 3 turns and the input winding is 12 turns. Wire dimension is uncritical.

The drain transformer, TR3, is wound on a ferrite toroid core from Ferroxcube (Philips). Trifilar 3 x 9 turns. Material 4A11, type TN 16/9.6/6.3.

Fig.1. Circuit diagram for 2.5MHz RF amplifier and filter.

The 2.5MHz band pass filter

The band pass filter has a flat response over 90kHz. The coils are all equal with an inductance of about 10 micorhenry.

These inductors are critical. If a single toroid Amidon T80-2 with about 40 turns is used for these inductors, IP3 is degraded by 10dB within the passband. Air core coils become very big for a similar Q but allow an higher IP3.

It seems reasonable to use the Amidon cores. Well outside the passband a much smaller voltage is present across the inductors so IP3 will gradually go from 30dBm within the passband to 40dBm for signals well outside the passband. Within the passband dynamic range is limited by the Delta44 saturation at -16dBm so IP3 = 30dBm means that third order intermodulation is at -90dBc with near saturating signals within the passband.

The frequency response of the 2.5MHz filter is shown in figures 2 and 3.

Fig.2. Frequency response of RF filter. Vertical is 3dB per division and horisontal is 20kHz per division.

Fig.3. Frequency response of RF filter. Vertical is 10dB per division and horisontal is 200kHz per division.

Dynamic range for in-band signals

When one near saturating signal is fed into one channel of the 2.5MHz to audio converter the spectrum looks like in figures 4 and 5. In the prototype there is one channel only so the spectrum is showing both upper and lower sidebands from mixing with 2.5MHz. When two channels of this kind are combined as I and Q, the noise floor will go down by 3dB. (actually it goes up by 3 dB and signals go up by 6dB for which linrad compensates)

Fig.4. Near saturating signal at 2.499MHz.

Fig.5. Near saturating signal at 2.509MHz.
When both signals are applied at the same time the amplitude has to be reduced by 6dB for the peak value of the signal to reach the same peak value as in figures 4 and 5. Such an intermodulation test is shown in fig. 6.

Fig.6. Two signals with equal amplitude, 2.409MHz and 2.509MHz. Each signal is attenuated by 6dB compared to figures 4 and 5 respectively but their combined amplitude is the same as in figs 4 and 5, 0.8dB from saturation.
The spectrum shown in fig. 6 is perhaps not very impressive because of the many visible spurs until one looks at the actual numbers on the dB scale. All the figures 4 to 8 are made with a bandwidth of 25Hz but the screen has only 1024 pixels for 46kHz so the S/N is actually higher than the figures indicate because peak height is lost on averaging in the x direction.

To give some idea about the source of nonlinearities figures 7 and 8 show the same two signals as fig 6 but with 11dB attenuation. For fig. 7 the attenuation is placed between the 2.5MHz to audio converter and for fig. 8 the attenuation is placed in front of the 2.5 to audio converter.

Fig.7. Same as fig. 6 but with 11.5dB attenuation between 2.5MHz to audio converter and Delta44.

Fig.8. Same as fig. 6 but with 11.5dB attenuation in front of the 2.5MHz to audio converter.
The strongest spur, third order intermodulation from the Amidon cores comes at 11 and 19kHz. This spur coincides with higher order spurs from the Delta44. Regardless if attenuation is placed in front of or behind the 2.5MHz to audio converter the level is the same, -105dBc so it is probably mainly caused by the signal source. Second order audio intermodulation, sum and difference frequencies are produced by the AD797 op-amps that drive rather low impedances to high voltage swings. The Delta44 mainly produces sums and differences of second harmonics.

When the system is used for weak signal communication the noise floor will be 15 to 20 dB higher than it is in figures 4 to 8 because the noise from the preamplifier should be dominating completely for a good system noise figure. The spurs shown in the screen dumps above will not be a problem but saturation will occur somewhere around 130dBc/Hz for signals within the passband. Outside the visible passband the dynamic range is much larger, it will probably depend on the mixers and amplifiers used to convert from other frequencies to 2.5MHz.

Fig.9. The second channel differs only in component numbering.