There are a few things that can cause "bleedover" to the opposite sideband. One is IMD - PureSignal can help with that if you choose to use it. The other is simply how various parameter selections are effecting the "sharpness" of your bandpass filters. Transmit filter "sharpness" depends upon selection of sample rate and DSP transmit buffer size. The higher the sample rate, the larger the buffer should be. At, for example, 192K sample rate, I'd recommend no less than a 4096 TX buffer size. One other parameter - select the BH-4 Window for maximum sharpness.
You also might want to check the lower frequency limit of your TX bandpass filter. Again, using 192K and 4096 buffer size as an example, I probably wouldn't go below 250Hz or so as the lower limit on the TX filter. In any case, you'll be able to view the results on the panadapter as you adjust.
Note that when in transmit mode, and NOT using the DUP function, the panadapter will show you your TX signal as it comes out of the software DSP - that will allow you to adjust the things I've described without seeing the effects of IMD. If you then click DUP, you will get to see the additional effects of everything that happens after the signal leaves the PC - presumably the additional effects will be primarily IMD.
(NR0V 31Oct14)
You are correct that Pure Signal cannot improve the starting waveform. When you increase the DSP buffer size, you are making the cutoff of the transmit bandpass filters sharper, thereby eliminating more energy in the opposite sideband. PureSignal reduces the Intermodulation Distortion (IMD) that is created in the analog amplifier chain. So, these are really two different things. You can choose your DSP buffer size to reduce the opposite-sideband energy to an acceptable level by looking at your transmit signal on the panadapter when NOT in 'DUP' mode. Then, you can look at the final result including IMD reduction using PS by receiving your own signal in the DUP mode.
As far as recommended buffer sizes, to an extent it depends upon your desires for suppression outside the transmit bandwidth. Also, it's a function of the sample rate you choose. When you double the sample rate, you have to double the DSP buffer size to get the same filter sharpness. As far as my personal choices, I use:
SampleRate = 48K; buffer >= 1024
SampleRate = 96K; buffer >= 2048
SampleRate = 192K; buffer >= 4096
SampleRate = 384K; buffer >= 8192
One more factor: There is now a choice of filter window function. The 4-term Blackman-Harris window, "BH - 4" will produce a filter with sharper cutoffs than the "BH - 7" window. BH-4 for sharp cutoffs and BH-7 for deep cutoffs. BH-4 was introduced in Jan 2014
(NR0V 18Apr14)