To sum it up: the only difference of this function to the urldecode function is that the "+" character won't get translated.

Hi everybody =) My name is Javier and I'm from Argentina.
I've had a little issue with latin characters like ñ","Ñ","á","é","í", etc.
They are not decoded with rawurlencode(), so I've made this:
<?php
function urlRawDecode($raw_url_encoded)
{
    # Hex conversion table
    $hex_table = array(
        0 => 0x00,
        1 => 0x01,
        2 => 0x02,
        3 => 0x03,
        4 => 0x04,
        5 => 0x05,
        6 => 0x06,
        7 => 0x07,
        8 => 0x08,
        9 => 0x09,
        "A"=> 0x0a,
        "B"=> 0x0b,
        "C"=> 0x0c,
        "D"=> 0x0d,
        "E"=> 0x0e,
        "F"=> 0x0f
    );
        
    # Fixin' latin character problem
        if(preg_match_all("/\%C3\%([A-Z0-9]{2})/i", $raw_url_encoded,$res))
        {
            $res = array_unique($res = $res[1]);
            $arr_unicoded = array();
            foreach($res as $key => $value){
                $arr_unicoded[] = chr(
                        (0xc0 | ($hex_table[substr($value,0,1)]<<4)) 
                       | (0x03 & $hex_table[substr($value,1,1)])
                );
                $res[$key] = "%C3%" . $value;
            }

            $raw_url_encoded = str_replace(
                                    $res,
                                    $arr_unicoded,
                                    $raw_url_encoded
                        );
        }
        
        # Return decoded  raw url encoded data 
        return rawurldecode($raw_url_encoded);
}

print urlRawDecode("%C3%A1%C3%B1");

// output:
// áñ

?>
For example, you have the character "ñ" encoded like this "%C3%B1". 
This is nothing more and nothing less than 0xc3 and 0xb1, 
they are binary numbers, (HHHH LLLL, where HHHH=High and LLLL=Low).
0xc3 = 1100 0011 (binary 8 bit word), 0xb1 = 1011 0001 (binary 8 bit word),
To convert a raw encoded character to ascii we have to make boolean operations
between this two operands (0xc3 and 0xb1), boolean algebra were defined by George  
Boole, we need to use them here. The first one we going to use is the 
logical OR ("|" or "pipe") and logical AND ("&" or "and person").

A logical OR implies the following truth table:
a b (a OR b)
0 0     0
0 1     1   (a OR b or Both, a and b, must be true to get a true result)
1 0     1
1 1     1

A logical AND implies the following truth table:
a b (a AND b)
0 0     0
0 1     0   
1 0     0
1 1     1   (Both a AND b, must be true to get a true result)

So, here we have to make a logical OR with both 0xc3 and 0xb1 HIGH nibble,
a nibble is a half byte (4 bits), so we have to make a logical OR between 
1100 (0xc) and 1011 (0xb), we going to get this: 1111 (0xf), then we have to make
a logical AND between both LOW nibble, 0011 (0x3) and 0001 (0x1), we going to get 
this: 0001, so, if we want to see the final result, we have to put HIGH and LOW 
nibble on his Byte position, like this: 1111 0001 (0xf1) and that is nothing 
more and nothing less than "ñ" (to check this out, try the following: print(chr(0xf1));).

This "<<" is a logical shift left, if we have this binary number 0001 (1) and we make this:
0001 << 2 we'll get 0100 (4) right bits are filled with 0's.

<?php
# Conversion example %C3%B1 to ASCII (0x71)
print(
    chr(
        (0xc0|0x0b<<4) | (0x03&0x01)
    )
);

// Output will be:
// ñ

// 1100 0000 OR 1011 0000 = 1111 0000 (0xf0)
// 0000 0011 AND 0000 0001 = 0000 0001 (0x01)
// 1111 0000 OR 0000 0001 = 1111 0001 (0xf1)

?>

PS: I'm so sorry about my english, I know, is horrible :P

Be aware that rawurldecode does not warn you in any way if the output is nonvalid UTF-8.
For example if the input passed to the function is just "%C5", then since C is 1100 in binary, and UTF-8 characters starting with 110 should be followed by another character, the result of rawurldecode will be just a single byte (with value \xC5) which is not a correct UTF-8.
Confront this with for example Javascript which will warn you about it:

JAVASCRIPT:

decodeURI("%C5")
URIError: URI malformed

decodeURIComponent("%C5")
URIError: URI malformed

unescape("%C5")
"Å"

PHP:
var_dump(rawurldecode("%C5"))
string(1) "▒"

php -v
PHP 5.3.6 (cli) (built: Oct  4 2012 10:19:07)
Copyright (c) 1997-2011 The PHP Group
Zend Engine v2.3.0, Copyright (c) 1998-2011 Zend Technologies
    with Suhosin v0.9.32.1, Copyright (c) 2007-2010, by SektionEins GmbH