Design Article
Comment
BrettS
In reviewing these articles, it appears that the only real advantage of this ...
ivovi
SNR is not useless, but should be properly treated. One of good measures for ADC ...
Software Defined Radio: Don’t Talk to Me about ENOBs, Part 2 of 2
Scott Kulchycki, Ph.D., Staff Engineer, High-Speed Signal Path Group, National Semiconductor Corporation
9/25/2010 2:28 PM EDT
Editor's Note: Part 1 introduces and defines key specs for SDR evaluation.
The Limitations of Traditional ADC Specifications
SNR, SFDR, and ENOB are measurements that consider the entire Nyquist zone of the ADC in response to a single-tone sine wave input. However, most real-world signals are not single-tone sine waves and the bandwidth of the signal being digitized in most systems is almost never equal to the Nyquist bandwidth. Many applications are designed to process non sine-wave signals with bandwidths much smaller than the entire input bandwidth, including:
To illustrate the limitations of traditional ADC specifications applied to these types of real-world systems and signals, consider a cable TV spectrum with four channels broadcast at 57 MHz, 63 MHz, 75 MHz, and 81 MHz (see Figure 4). Because the cable TV spectrum can range up to 1.1 GHz, the ADC sampling rate must be at least 2.2 GSPS. Assuming we want to receive a channel at 69 MHz, the best measure of system performance is the smallest channel power that can be received at that frequency in the presence of system noise and in the presence of the adjacent channels.
Part 2 of this piece details the limitations of traditional specs for SDR, and which ones really matter. Download the PDF here (no registration, no funny business, just a link to the article)
The Limitations of Traditional ADC Specifications
SNR, SFDR, and ENOB are measurements that consider the entire Nyquist zone of the ADC in response to a single-tone sine wave input. However, most real-world signals are not single-tone sine waves and the bandwidth of the signal being digitized in most systems is almost never equal to the Nyquist bandwidth. Many applications are designed to process non sine-wave signals with bandwidths much smaller than the entire input bandwidth, including:
- Cable TV (6/8 MHz channels in a 1.1 GHz input bandwidth)
- Satellite TV (typically 36 MHz channels in a 500 MHz input bandwidth)
- Multi-carrier / multi-standard basestations (as small as 200 kHz channels in a 20 MHz bandwidth)
- Oscilloscopes (input bandwidth can be 10% or less of the Nyquist bandwidth)
- Weather radar (received data as low as 10% or less of Nyquist bandwidth)
To illustrate the limitations of traditional ADC specifications applied to these types of real-world systems and signals, consider a cable TV spectrum with four channels broadcast at 57 MHz, 63 MHz, 75 MHz, and 81 MHz (see Figure 4). Because the cable TV spectrum can range up to 1.1 GHz, the ADC sampling rate must be at least 2.2 GSPS. Assuming we want to receive a channel at 69 MHz, the best measure of system performance is the smallest channel power that can be received at that frequency in the presence of system noise and in the presence of the adjacent channels.
Part 2 of this piece details the limitations of traditional specs for SDR, and which ones really matter. Download the PDF here (no registration, no funny business, just a link to the article)
Navigate to related information


Frank Eory
9/27/2010 7:29 PM EDT
Great follow-up to Part 1, and a nice job explaining which specs matter in SDR and why.
Sign in to Reply
ivovi
10/13/2010 8:09 AM EDT
SNR is not useless, but should be properly treated. One of good measures for ADC noise performance is dBFS/Hz, which is power ratio of full-scale harmonic signal to noise density. Consider this article (sorry not in english), but pictures and equations are clear.
Note that this is status quo of 2004 year, nowday ADCs are about 4 dB better:
http://www.elektrorevue.cz/clanky/04050/index.html
Sign in to Reply
BrettS
10/28/2010 2:30 PM EDT
In reviewing these articles, it appears that the only real advantage of this technology is hardware footprint, but it still works analyzes signals serially except now the issue is noise and distortion comes into play. So when is the next technology coming that can take a signal and process it in parallel?
Sign in to Reply