Q1Digital Signal Processing
Question
Q.1. Compare direct form I and direct form II realization of IIR systems.
Answer
Direct Form I implements the numerator (zeros) and denominator (poles) sections as two separate cascaded delay chains, needing 2N delay elements, while Direct Form II combines them into a single shared delay chain, needing only N delay elements (canonical form).
For an IIR system with transfer function H(z) = B(z)/A(z) = (b0 + b1z^-1 + ... + bMz^-M)/(1 + a1z^-1 + ... + aNz^-N), two standard structures realize this transfer function.
Direct Form I: implements the difference equation directly as two separate sections in cascade — first a feedforward (all-zero, FIR) section applying the numerator coefficients bk to delayed versions of the input x(n), followed by a feedback (all-pole, IIR) section applying the denominator coefficients ak to delayed versions of the output y(n). This requires a total of M delay elements for the input section plus N delay elements for the output section (M+N total delays), since the two delay chains (for x and for y) are entirely separate and not shared.
Direct Form II (canonical form): rearranges the same transfer function by swapping the order of the pole and zero sections and observing that, since they are now in series, the same intermediate delayed signal values can be shared between the two sections rather than maintaining separate delay chains for input and output. This reduces the total number of delay elements to max(M,N) (typically N when M ≤ N), making Direct Form II the 'canonical' realization since it uses the minimum possible number of delay elements/memory for a given order transfer function, at the cost of the internal signal (the shared delay-line values) potentially having a larger dynamic range than in Direct Form I, which can be more prone to overflow/quantization issues in fixed-point implementations.
Comparison: Direct Form I is more robust to coefficient quantization and overflow (since the pole and zero delay lines are separate, an overflow in one section does not directly corrupt the other), but uses more memory (M+N delays). Direct Form II uses minimal memory (canonical, max(M,N) delays), which is valuable in memory-constrained embedded/DSP hardware implementations, but its shared internal delay line can experience larger intermediate signal swings, making it more susceptible to fixed-point overflow and requiring more careful scaling in practical hardware implementations.