You could have designed state of the art positional encoding

If you're interested in positional embeddings for Transformers, check out this repo - https://github.com/gazelle93/Attention-Various-Positional-En... - it implements various popular ones.