What is the structure in this question? "it's difficult knowing which one to choose."

This sentence is grammatical, but it's complex, and has been twisted about a bit.

• It's difficult knowing which one to pick.

There are three verbs in the sentence (be, know, and pick), so there are three clauses. The main predicate is be difficult, a predicate adjective, which uses be as an auxiliary, like all predicate adjectives and nouns. The other clauses are subordinate to the main clause.

If is difficult is the main verb phrase, what's it's subject? Well, everything else in the sentence.

(Except it. The it is a meaningless dummy, introduced by the syntactic rule of Extraposition.)

Knowing which one to pick is also a clause, a gerund clause with knowing as the gerund main verb of the clause. But every English clause needs an English subject, so whose knowing are we talking about? It's not clear.

Anybody at all, everybody, somebody, whoever you like. It's indefinite who the subject is (except it does have to be a Who, not a What, because that's required for the subject of know), and since it's not there, we just don't know. But we can tell that the subject is some indefinite human, and not somebody mentioned in the conversation, so we'll just call it Indef, which of course, is not pronounced. Like apostrophes.

So the subject of is difficult is the clause

• Indef knowing which one to choose

and together the sentence is pronounced

• Knowing which one to choose is difficult.

This is the sentence that's left when you unwind Extraposition. That moves a heavy subject like a gerund clause to the end of the sentence where it's easier to process, and leaves behind a dummy It (normally contracted to It's when the predicate is an adjective, as here) behind to fool us into thinking there's a subject still there so we can get to the verb fast. There are a lot of rules in English that do this, in various ways to various kinds of construction, but always in order to put heavy stuff at the end of the sentence.

But we're not done. The gerund subordinate clause knowing which one to choose has a subordinate clause itself: which one to choose is a Wh-infinitive clause, and once again its subject is missing, and once again it's indefinite.

Somebody (indefinite) is to choose something (also indefinite, so it may be somebody instead) from a limited set of things (or people). That's the sense of which one instead of what, which could be plural and isn't limited in scope.

Wh-infinitives are very restricted and have extremely peculiar syntax. The subject of choose is Indef, but the object is which one, moved to the front as a relative pronoun. One in which one is the indefinite singular pronoun, standing for some noun that is understood in the discourse. We don't know what (or who) is being chosen, nor by whom, but the speaker is relying on the addressee knowing what noun one stands for. This is not "indefinite" in the same logical sense -- we still don't know which one, but we know what kind of ones we're talking about.

The Wh-infinitive is the direct object of knowing, so that makes it a noun clause, a complement of know. The gerund clause itself is the subject of is difficult, so that makes it a noun clause, too. And of course there's Extraposition. Pretty complicated structure for such a small sentence.

Just an addendum to John Lawler's excellent answer: your sentence

[1] It's difficult knowing which one to choose.

is arguably only marginally idiomatic in its intended meaning, which I take to be epistemological ('in this case one is unsure which one to choose') rather than emotive ('the fact that one knows which one to choose exerts an emotional toll on one.'). If the surrounding context makes it clear that it is meant emotively, then it is fully acceptable. However, the emotive reading on the whole seems unlikely for this sentence, and it would require very special context (Lambie gave one in the comments to the original question).

Assuming the meaning is epistemological, then, to make [1] more idiomatic, the gerund-participial clause knowing which one to choose should be replaced by the to-infinitival clause to know which one to choose:

[2] It's difficult to know which one to choose.

Once this replacement is made, John Lawler's analysis goes through unchanged (of course, where he says 'the gerund subordinate clause knowing which one to choose…', one should now put 'the to-infinitival subordinate clause to know which one to choose…').

As John Lawler said, the OP's sentence has a 'basic' counterparts, i.e. a counterpart which is not an extraposition. The sentence [2] has such a counterpart as well:

[1a]  Knowing which one to choose is difficult.
[2a]  To know which one to choose is difficult.

In this form, both are idiomatic.

Discussion

Why is it that [1] (the extraposed form of [1a]) is less acceptable than [2] (the extraposed form of [2a])? All I can say is what CGEL says, namely, that 'gerund-participials extrapose less readily and generally than content clauses and infinitivals.' Here is the relevant section in full (p. 1407):

7.2 Further cases of extraposition

(a) Extraposition of gerund-participial subjects

Extraposition can apply with gerund-participials as well as content clauses and infinitivals:

[11]   i  a.  Complaining would be no use.                       b.  It would be no use complaining.
        ii  a.  Getting Ed to agree to our proposal       b.  It will be no problem whatsoever
                 will be no problem whatsoever.               getting Ed to agree to our proposal.

We did not include this under the central case of extraposition because gerund-participials extrapose less readily and generally than content clauses and infinitivals. Compare, for example:

[12]   i  It was stupid to tell my parents /^?telling my parents.
         ii It would make things worse to call in the police /^?calling in the police.

Here the extraposed infinitivals are impeccable, but the gerund-participials are at best very marginal, though the basic versions are quite unproblematic (Telling my parents was stupid; Calling in the police would make things worse). Gerund-participials with an overt subject, especially a non-pronominal one, are particularly resistant to extraposition: Kim and Pat getting married had taken us all by surprise, but not *It had taken us all by surprise Kim and Pat getting married. Note that gerund-participials differ from infinitivals in that they can undergo subject-auxiliary inversion (cf. Will getting Ed to agree to our proposal be a problem?), so extraposition is not obligatory in this context.²¹ Gerund-participials in complement function are the most NP-like of subordinate clauses, and the fact that they are less amenable to extraposition than the other subordinate clause categories is a further manifestation of this.

^{21There is, however, one item that takes a gerund-participial—namely, worth—that enters into the pattern shown in [10] [these were examples of verbs or verbal idioms which enter in extraposed contstructions without any basic counterparts—more precisely, they occur with dummy it as the subject and a declarative content clause or infinitival in post-verbal position without an equivalent version where the subordinate clause appears as subject. On example is fall. We can say it fell to me to notify her family but not *To notify her family fell to me. Other examples are appear, be, chance, come about, fall out, happen, seem, strike, transpire, turn out, fall, remain and, in the passive, decide, hope, intend]. We have, for example, In discussing the future it is also worth considering the impact on Antarctica of human activities elsewhere on the globe, but not *Considering the impact on Antarctica of human activities elsewhere on the globe is worth.}

The major corpuses of English to which I have access confirm that it's difficult knowing is rare, much more so than it's difficult to know.

If one searches Google NGrams for the phrase "it's difficult knowing," (or "it is difficult knowing"), one gets zero hits, whereas the phrases "it's difficult to know" and "it is difficult to know" do get hits.

The same result is obtained if one searches the Corpus of Contemporary American English (COCA) or the British National Corpus (BNC).

If one searches google books for "it's difficult knowing," one gets hits like this:

[It's difficult] knowing that they are doing the same “family” [activities] we enjoyed while married, including [going to the same] vacation places. (source)
It's difficult knowing the hardest part is yet to come. (source)
It's difficult knowing, however, that for every detail I get right, another detail is likely wrong. (source)
It's difficult knowing in advance of a decision precisely what information will be most valuable. (source)

Clearly, in these examples, the phrase means that knowing these things exerts an emotional toll on the knower, which is different from the more epistemological meaning which it is supposed to have in your sentence.

True, google books does return a few hits where the phrase is used with the same meaning as in your sentence:

It's difficult knowing if I'm making progress or not. (source)
It's difficult knowing who to trust in those circles. (source)
Sometimes it's difficult knowing where the line is, whether I will do more harm than good. (source)
It's difficult knowing what to take seriously these days. (source)

But if you search for it's difficult to know, you will see that that phrase is much more common.