How best to structure database with multiple possible join tables

Question

I'm trying to work out how best to structure some database tables so I can ensure my response entities are as coherent as possible.

Basically say I have employees and I have different documents, that employee can have create a document of each type, however, those documents could be any one of 4 possible types of documents and the information and structure of those documents is completely different from one another... How would be best to structure that to get a nice response?

Of course I could do 4 many-to-one joins with a table called employee with 4 document tables called document_a, document_b and so on joined on via an employee_id column... but this would be somewhat messy in a JSON response as it would be akin to:

{
    {
        "firstName": "first",
        "lastName": "last",
        "documentAs": [
           {
             "field1":"test",
             "field2":"test"
           }
        ],
        "documentBs": [
           {
             "field3":"test1",
             "field4":"test2"
           },
           {
             "field3":"test3",
             "field4":"test4"
           }
        ],
        "documentCs":[],
        "documentDs":[],
    }
}

Ideally though, I would want the JSON response to be closer to the following:

{
    {
        "firstName": "first",
        "lastName": "last",
        "documents": [
           {
             "documentType": "documentA",
             "field1":"test",
             "field2":"test"
           },
           {
             "documentType": "documentB",
             "field3":"test1",
             "field4":"test2"
           },
           {
             "documentType": "documentB",
             "field3":"test3",
             "field4":"test4"
           }
         ]
    }
}

Is there any realistic database structure that could allow me to do get this kind of response as I can't think one of due to their different table structures? Of course I could do the 4 many-to-one joins and manipulate the data afterwards into the structure I want but that would be bad performance-wise and resrict filtering and possibly other functionality. To be clear, the fields in each document would be fixed, not dynamic, but unique and different to each document type.