Given a Turing machine M and a string w, you can show that the following language is accepted by some LBA:
L = { x#y | x is a computation trace of M accepting w and y is any string }
Intuitively, this can be checked by an LBA by having it do the following:
- Reject if x is not syntactically correct.
- Reject if x doesn't start off in a correct initial configuration.
- Reject if any step of the computation trace is incorrect.
- Reject if x is a trace showing M rejects w.
- Otherwise accept
It's possible for a TM to construct a description of the LBA that does this.
If M accepts w, then this language is infinite and so the LBA will accept infinitely many inputs. If M does not accept w, then this language is empty. Therefore, if a TM could decide whether the LBA had an infinite language, it could decide whether M accepts w, contradicting that this is impossible.
Hope this helps!