Circuit noise analysis and optimization, Part 3a

The word “noise” suggests a range of electrical disturbances. The source of some of these disturbances is inherent to analog circuit components. such as amplifiers and converters. To know the lowest signal level a circuit can reliably process, a designer has to calculate the circuit noise.

This can be complex, as it involves deriving and solving many equations. Before writing equations, one has to identify and account for uncorrelated and correlated noise sources. This article will examine special situation where both correlated and uncorrelated noise exist, and offer optimization suggestions for the circuit. In this way, a designer can build circuits that function effectively with optimum noise performance.

My noise articles published at Planet Analog in October 2008 (links below, at the end) covered many of the basic concepts, but left room for many questions about noise calculation and noise optimization of various circuits. This article responds to these issues. It may be helpful to review the previous articles as they are the basis for this new article, and many references will be made to the previous material.

Beginning the analysis
To obtain the best noise performance, one has to limit the bandwidth and calculate noise over the noise effective bandwidth (NEB). One bandwidth-limiting scheme uses a simple low-pass filter in, or after, the signal-conditioning stage.

Noise bandwidth is wider than the signal bandwidth regardless of filter order of the system. A signal has fallen to 0.707 times of its original value at its –3 dB frequency (f0 ), but noise bandwidth for a first-order filter extends to 1.57 f0 . This means that white noise is passed as if the filter were a brick-wall type with a cutoff frequency 1.57 times as large. Multiplication factor for higher-order (N) low-pass filters are shown in Table 1 .

Table 1. Multiplication factors give noise equivalent bandwidth for filters of order N

(Click on image to enlarge)

To calculate the noise equivalent bandwidth of an arbitrary circuit, such as the band-pass filter shown in Figure 1 , a designer could use Equation 1 , which integrates the magnitude of the signal gain over frequency, and divides it by maximum signal gain.

Figure 1: Band-pass filter

(Click on image to enlarge)

(Equation 1)

(Click on image to enlarge)

An easier way to calculate the circuit noise is to use simulation software such as PSpice. This band-pass filter has cut-off frequencies of (1/2π ×R3 ×C1 ) and (1/2π ×R1 ×C21 ). It is important to have an accurate op amp model that models the voltage and current noise densities. The process of noise calculation is the same as hand calculation, but the software does the data crunching.

To find NEB, select “AC sweep” from the “analysis setup” popup menu and click on the “noise enabled” button. The first step is to find the maximum gain, Figure 2 .

Figure 2: Maximum gain of circuit shown in Figure 1.

(Click on image to enlarge)

Next, the summation operators can be used to find NEB, Figure 3 . The NEB is where the upper part of the graph flattens out, roughly 170 kHz in this example.

Figure 3: Noise equivalent bandwidth (NEB) of circuit shown in Figure 1

(Click on image to enlarge)

Pspice also calculates other pieces of information that a designer needs to study. For example, it has already gathered the necessary data to display the input and output noise graphs for the circuit shown in Figure 1. This is done by selecting “enable noise” in the setup menu. The output noise result is shown in Figure 4 .

Figure 4. Output noise of circuit shown in Figure 1.

(Click on image to enlarge)

Knowing the noise bandwidth allows a designer to find the noise over this bandwidth by using probe and the s operator, Figure 5 . Place one cursor at low frequency (e.g., 1 Hz). Right click and place a second cursor at 170 kHz circuit's noise bandwidth). The noise over this bandwidth can be read from “dif” in probe cursor.

Figure 5: Total output noise of circuit shown in Figure 1 integrated over NEB.

(Click on image to enlarge)

The above approach assumes that all of the circuit's noise sources are correctly identified and are configured for optimum noise performance. Pspice does not know the correct configuration and component values, which must be optimized by the circuit designer. Is the above circuit optimized for noise? Are all noise sources identified and accounted for?

(End of Part 3a ; Part 3b will look at an actual circuit example, with calculations.)

About the author
Reza Moghimi is an applications engineering manager for the Precision Analog Products group at Analog Devices, Inc. He holds an BSEE and MBA from San Jose State University (SJSU), CA. In addition to Analog Devices, Reza has worked for Raytheon Corp., Siliconix Inc, and Precision Monolithic Inc. (PMI). He enjoys traveling, music and soccer.

Previous parts of “Understanding Noise Optimization in Sensor Signal-Conditioning Circuits”:

  • Part 1a: click here
  • Part 1b: click here
  • Part 2a: click here
  • Part 2b: click here

1 comment on “Circuit noise analysis and optimization, Part 3a

  1. oijsoifjdsoifj
    September 28, 2015

    Looking to buy a home? It's better to be on a “Way” than a “Street,” pick a female real-estate agent and try to be close to a Starbucks.
    That's the advice of Spencer Rascoff, CEO of, who collected statistics from his site's database of 110 million homes to find trends in real-estate pricing. Along with Zillow economist Stan Humphries, he has written “The New Rules of Real Estate” (Grand Central), out Tuesday. Some of his findings:
    The Starbucks effect. Take two identical homes sold in 1997. One near Starbucks would have sold for an average of $137,000, while the same home without a Starbucks would have sold for $102,000. Fast-forward 15 years: the average US home appreciated 65 percent to $168,000, but the property next to Starbucks skyrockets 96 percent to $269,000.
    All renovations are not created equal. The greatest return for your investment is a mid-range bathroom remodel, a $3,000 job that returns $1.71 for every dollar spent. The worst home improvements for value are kitchen remodeling and finishing a basement. A top-of-the-line kitchen reno will cost you $22,000, and you'll only get about $0.51 back for every $1 you spend.
    Use the right words in a listing. Avoid “unique,” “TLC,” “investment” and “potential” — these could lower sale prices by as much as 7 percent. But words like “luxurious” for bottom-tier homes and “captivating” for top-tier homes could add 8.2 percent to your home's value. Longer, more-detailed listings often sell for more.
    “When” is as important as “how much.” In New York, the worst time to sell is the second week of December (listings sold for 2.8 percent less than average). The best time is March, when homes sold faster and for 2 percent more.
    Seven is an unlucky number. Homes with “777” as their address sell for 2.1 percent less than their estimated value; house numbers that just include 777 (such as 17779 Main St.), sell for 1.8 percent less. Oddly, houses with just 7 as their number sell for 1.8 percent more than the estimated sale price.
    Psychological pricing works. Listings with a nine in the thousand digit ($450,000 vs. $449,000) sell anywhere from four days to a full week faster.
    Female agents tend to sell homes faster and for higher prices.
    What's in a name? A lot of cash, according to Zillow's data. Homes on named streets tend to be 2 percent more valuable ­nationwide than numbered ones (unless you're talking about New York City, where it's a wash). But Main Street homes garner 4 percent less than America's median home value. Street names with Lake or Sunset will sell upwards of 16 percent higher. Suffixes also matter. Avoid “Street,” which has the lowest home values of $183,120 nationally, and find a “Way,” which has the highest home values averaging around $312,000.


Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.