I'm trying to write my own solution for a BFE problem https://bigfrontend.dev/problem/implement-atob

Basically, writing a polyfill for the atob function available in javascript

I tried to understand the base64 encoding/decoding process and came up with the following implementation that works fine for most inputs but some of the test cases are failing

function myAtob(encoded) {
// All base64 charset
const base64 = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";

let temp = ""; // used for storing all characters to binary form
let decoded = "";

for (let i of encoded) {
    // converts each encoded character to binary and pads if required
    temp += base64.indexOf(i).toString(2).padStart(6, 0);
}

for (let i = 0; i < temp.length; i += 8) {
   const num = temp.slice(i, i + 8); // make groups of 8 bits
   const ascii = parseInt(num, 2); // convert 8 bits to number
   if (ascii > 30 && ascii < 123){
       // I used above 'if' for ignoring the unsupported characters
       decoded += String.fromCharCode(ascii); // append string equivalent
   }
}

return decoded;
}

I tried to debug and I believe it's happening because some unsupported codes or escaped characters are getting part of my output messing up the test cases.

Example when using the window.atob function,

atob("A===") throws error
atob("AA==") returns '\x00'

These are different from what my function returns. Maybe the padding character = is messing up something.

The following are test case results:

Can someone help me with fixing the logic and handling of edge cases?