Non-linear digital rank filtering of input signal values

A method of non-linearly digitally filtering K input values to produce an output value corresponding to an Rth greatest value among said K input values with K≧R≧1, comprising the steps:

• a) providing said K input values, each respectively in a signed binary fixed-point numerical representation having plural numerical bits in order of a respective bit significance thereof from a most significant bit to a least-significant bit in succession, wherein each one of said numerical bits has a bit value of only 0 or 1, said bit significance diminishes by a factor of ½ from each one of said numerical bits to a next one of said numerical, bits in succession from said most-significant bit to said least-significant bit, and said numerical representation of each one of said K input values respectively may further optionally include a sign bit;
• b) if said numerical representation does not include said sign bit then skipping the remainder of this step b), otherwise if said numerical representation includes said sign bit then evaluating said sign bits of said K input values as follows:
• b1) if all of said input values are non-negative values or all of said input values are non-positive values then skipping the remainder of this step b);
• b2) if there are at least R non-negative values and at least one negative value among said K input values then setting an output sign indicator as non-negative and setting all of said numerical bits of all of said at least one negative value to 0;
• b3) if there are fewer than R but at least one non-negative value among said K input values then setting an output sign indicator as negative and setting all of said numerical bits of all of said at least one non-negative value to 0;
• c) after said step b), selecting a respective current selected bit among said numerical bits of each of said K input values beginning with said most-significant bit and progressing successively with said bit significance diminishing toward said least-significant bit;
• d) after said step c), determining a respective output bit of said output value for said respective current selected bit of said K input values as follows:
• d1) if said numerical representation does not include said sign bit or if at least R of said K input values are non-negative values, then setting said respective output bit to 1 if said bit value 1 occurs at least R times as said current selected bit among said K input values and otherwise setting said respective output bit to 0;
• or
• d2) if fewer than R of said K input values are non-negative values, then setting said respective output bit to 0 if said bit value 0 occurs at least R times as said current selected bit among said K input values and otherwise setting said respective output bit to 1;
• e) after said step d), revising said numerical bits following said respective current selected bit of said K input values, to the extent necessary as follows, by setting any and all of said numerical bits following said respective current selected bit to be equal to said respective current selected bit respectively in any and all of said input values of which said respective current selected bit does not correspond to said respective output bit;
• f) repeating said steps c), d) and e) successively with successive ones of said respective current selected bit with said bit significance thereof successively diminishing in successive cycles of said steps c), d) and e), until all of said output bits of said output value have been or can be determined;
• and
• g) outputting said output sign indicator to indicate a sign of said Rth greatest value and outputting said output value as a value of said Rth greatest value.