This has nothing to do with compression of any kind. You need to consider resolution and pixel dimensions in your source images. Think of them as pixel grids.
Here you're apparently placing a low-res image into a high-res one. When you rasterize (commit) the smart object, it will therefore get upsampled to align with the base pixel grid. Hence the softening.
If you had just copy/pasted it in, it would be much smaller - because in terms of pixels, it is much smaller. So one pixel grid would just snap into the other.
Smart objects is a way to temporarily disregard pixel size, and honor relative print dimensions instead. But ultimately, everything in Photoshop has to go into the base pixel grid of the document.
I seem to recall years ago when using the Place command you were given opportunity to choose a resolution value in the dialog that appeared, but that was removed long ago. No idea why, but it was long before embedding or linking and Smart Objects were invented, so maybe that had something to do with it.
You still get that when opening PDFs (and EPS), maybe that's what you're thinking of?
Anyway, on closer look, the kitten doesn't appear upsampled as much as simply misaligned to the base pixel grid. Or maybe it's just a poor quality preview, and the second screenshot is how it really looks.
What are the original pixel dimensions of the kitten, and of the base document? That's number of pixels - not in/cm or ppi, which don't matter here.
actually i'm placing the kitten (High Resolution) to a 85x85 (Low resolution). I'm trying to find a way to use the original preview instead of the final image.
Fixed it. opening it in another project and using free transform didn't sample it.