What is the reconstruction filter design for a DAC-based RF signal generator?

The reconstruction filter design for a DAC-based RF signal generator removes the DAC's output images (aliases) that appear at multiples of the DAC sample rate. The DAC produces replicas of the desired signal at f_out ± N × f_sample (where N is any integer). The reconstruction filter (also called an anti-imaging filter or smoothing filter) is a low-pass or bandpass filter that passes the desired signal and rejects all image frequencies. Design considerations: passband (the filter must pass the desired output frequency range with minimal insertion loss and group delay variation; for a baseband output: the passband extends from DC to the maximum output frequency; for an RF output using a higher Nyquist zone: the passband is the desired RF band), stopband (the filter must reject the first image, which is located at f_sample - f_out (for the first Nyquist zone); the required rejection depends on the system's spectral purity requirement; typical: 40-60 dB rejection of the first image), transition band (the frequency range between the passband edge and the stopband edge; narrower transition bands require higher-order filters (more components, more insertion loss); for a direct RF DAC: the signal may be close to f_sample/2, requiring a very sharp filter), and filter topology (Butterworth: maximally flat passband, moderate rolloff; Chebyshev: steeper rolloff but passband ripple; Elliptic: steepest rolloff but passband and stopband ripple; for reconstruction: elliptic filters are often preferred because the narrowest transition band is critical). The sinc rolloff: the DAC's zero-order hold creates a sinc(x) frequency response that adds rolloff within the first Nyquist zone (approximately 3.9 dB at f_sample/2). This rolloff can be corrected digitally in the FPGA (sinc^-1 pre-emphasis) before the DAC, or: accepted and compensated in the system calibration.