Switched attenuator settings change output frequency in oscillator

Hi, in my signal generator http://qrp.gr/rfgenerator/ I have included a 50 ohm output switched attenuator. However, when I switch the 20db, the generator frequency changes a bit (500HZ or so). When I switch off the 20db and switch on the 10db, it changes less. If I switch the 3db only, it changes much less. If I keep the 20db attenuator on, then the 3 or 10db attenuators have almost no effect in frequency.
Is that normal?
Shall I include a 20db fixed attenuator always connected to the output, to tackle the problem?
This will also decrease the output level though.

Always an attenuator placed at the output of an oscillator helps for reducing frequency pulling (higher attenuator value, lower the frequency pulling).
The output level will be attenuated but this could be compensated by an amplifier.

You need more isolation beteee the tuned circuit and the output. The 20 dB pad is suplying this; assuming ideal about40dB, 20dB in the forward path and 20dB again in the reflected path.
A common base ampifier using a good HF transistor at the output would easily give this level of isolation over the HF range without the penalty of a 20dB drop in output power. If power level is not important, then stick with the pad.

The one drawback of the attenuator is that the wideband signal to noise ratio gets worse, i.e. the phase noise far away from the carrier gets 20dB worse. A friend of mine designs ultra low noise oscillators and runs them with high power levels of > 1W within the oscillator loop, for this exact reason: you can't lower the wideband thermal noise level, but you can increase the carrier level. His oscillators are the new world record test devices used at R&S to demo their phase noise analyzers.

But if you reduce the oscillator power by an attenuator, no amplifier down the signal chain can improve (restore) the wideband phase noise again.

Yes, using an attenuator at the oscillator output AND if the output level is low the signal to noise ratio will degrade, but in most of the oscillator designs if you do a trade off between frequency pulling and noise, the first always win.
In this regards, to design a high power oscillator is a good option, but generally is hard to get.

The same is in RF mixer design. There, more important than mixer noise figure is to get the best mixer linearity.

a fixed gain buffer amp at the output would help. Maybe with a 10 db pad or so in front of the buffer amp. You want to isolate your load from the generator's oscillator.
If it were microwave frequencies, you could use a ferrite isolator

here is an amp with a 57 db reverse isolation:
http://www.skyworksinc.com/uploads/d...ts/201518A.pdf

Thanks but way out of the frequency range of this HF generator.
Adding another stage for isolation and more power output is desirable, but not at the expense of increasing the harmonics to carrier radio.
Any such designs to try are appreciated (discrete preferred).