Skip to content
TESA Open Knowledge

Hands-on: cleaning a live signal with a filter

Module 4 — Signal analysis · Slides: slides.md · Module overview · Course page

Fill five points in s08_filters.py to build a filter by name with keyword arguments, read the acceleration magnitude, feed it into filt.update, and plot raw against filtered, then measure the result as noise down %, comparing three filters on the same signal.

Objectives

Section titled “Objectives”

By the end of this lesson, you will:

  1. Fill the five points in practice/s08_filters.py until the two charts clearly differ (the lower line is smoother than the upper one), and EMA, Median and Kalman1D can all be selected in the dropdown without an error.
  2. Record the noise down % of three filters on the same signal, and create a situation where Median clearly beats EMA, explaining why.
  3. Explain why the single line y = filt.update(x) works for every filter, and why noise down % is measured from frame-to-frame jitter.

Before you start

Section titled “Before you start”

You’ve been through lesson 4.1, and know the personalities of EMA, Median and Kalman1D. Keep your learning log ready to note the noise-down % for each filter.

Concepts

Section titled “Concepts”

The whole file reads as one sentence: read a raw value → feed it into a filter → plot raw against filtered → measure how many percent noise was reduced, looping every 60 ms. The filter is created in a “create once” step, through make_filter(name), which turns a name into an object. EMA is already given as an example — we fill in return dsp.Median(window=MED_WINDOW) and return dsp.Kalman1D(q=KAL_Q, r=KAL_R), which must always be keyword arguments. If you forget these two, the app still runs with EMA, but selecting Median or Kalman makes the function return None, which crashes when update is called.

read_raw() reads sensors.bmi270.acceleration() and collapses it into a magnitude, mag = (ax*ax + ay*ay + az*az) ** 0.5 (at rest, about 9.8), multiplied by 10 to show clearly on a 0..250 graph, since this lesson’s filters are one-dimensional, so the vector must be turned into a scalar first. Radar range mode reads sensors.radar_range() in centimetres, wrapped in try/except OSError. The heart of this lesson is the single line y = filt.update(x), which never needs to know what kind of filter filt is, because every one shares the same API. If you forget to fill it in (y = x), the two graphs will match exactly — a simple test of whether you’ve filled it in correctly.

We measure smoothness from frame-to-frame jitter: the sum of |x − x_prev| for the raw line versus the filtered line, then red = (1 − filt_jit / raw_jit) × 100. With a real sensor, we have no answer key for what the true value is, so a reduction in jitter is a straightforward indicator. Success isn’t “the line looks nice” — you should be able to say why Median beats EMA when a spike hits, and why a smaller α gets smoother but slower to follow.

Worked example

Section titled “Worked example”

s08_filters_full.py adds sliders to tune parameters live, depending on the filter type (α for EMA, window for Median, r for Kalman), a noise-down bar that changes colour by threshold, and a Freeze button to pause the picture and compare the spike against the smooth line.

File What this file teaches
examples/s08_filters_full.py Cleaning a signal with DSP filters (full version)

Practice

Section titled “Practice”

The 5 # TODO: comments are at lines 46 (Median), 49 (Kalman1D), 99 (acceleration magnitude), 144 (filt.update), and 148 (plotting the two graphs). Fill in one at a time, then move the board and switch through all three filters. If the two graphs never differ, check point 144 first.

Practice file Topic
practice/s08_filters.py Cleaning a signal with DSP filters (the fill-in-the-code version)

Solution

Section titled “Solution”

Open the solution after trying on your own at least once, and read how to use the solutions first.

Solution Pairs with
solution/s08_filters.py practice/s08_filters.py

Check your understanding

Section titled “Check your understanding”

The same questions are in quiz.yaml for automated checking.

  1. You’ve filled in the file, but the Filtered line matches the Raw line exactly, no matter which filter you choose. Which point is still empty? (single choice · objective 1)

    • a) return dsp.Median(window=MED_WINDOW)
    • b) mag = (axax + ayay + az*az) ** 0.5
    • c) y = filt.update(x)
    • d) raw_chart.value(clamp(x))
    Solution

    c — the placeholder y = x means the value drawn on the lower line is the raw value. The two graphs only differ once the value genuinely passes through filt.update.

  2. You write return dsp.Median(5) at point 1, then select Median. What happens? (single choice · objective 1)

    • a) You get the 5-value window you wanted
    • b) It throws TypeError, because window is a keyword-only parameter
    • c) You get a 6-value window
    • d) The filter becomes EMA
    Solution

    b — dsp’s filter parameters are keyword-only. You must write dsp.Median(window=5) — passing it positionally is rejected immediately.

  3. You tap the board hard once, creating a spike. What would you expect to see on the Filtered line for EMA versus Median? (single choice · objective 2)

    • a) Both remove it equally well
    • b) EMA smooths it into a soft bump that’s still visible; Median votes the spike away until it’s barely visible
    • c) EMA removes it entirely; Median smooths it into a bump
    • d) Both flatten the line to zero
    Solution

    b — EMA weighs the outlier into the result partially; Median picks the middle value, so it ignores a lone outlier entirely. This is the difference between “smoothing” and “voting it out”.

  4. raw_jit = 400 and filt_jit = 100. What’s the noise down %? (single choice · objective 3)

    • a) 25%
    • b) 75%
    • c) 100%
    • d) 400%
    Solution

    b — red = (1 − 100/400) × 100 = 75%. The filtered line’s jitter is down to a quarter of the raw line’s.

Lab

Section titled “Lab”

The MVP for lessons 4.1–4.2: a filter visibly improves a noisy sensor signal — the Filtered line is clearly smoother than Raw, confirmed with a noise-down % figure.

  • All five points in the practice file are filled in, and it runs on the emulator or the board.
  • Switch through all three filters on the same signal, and note each one’s noise-down % in your learning log.
  • Create a spike (tap the board hard once, or use the board’s radar), and explain why Median beats EMA.
  • Be able to explain which filter you’d choose, why, and what the smoothness-versus-response trade-off is.

Going further

Section titled “Going further”

In the next pair of lessons (4.3–4.4), we’ll look at the same signal in the frequency domain with FFT.

Next lesson: lesson 4.3 — FFT and the frequency domain: bins, Nyquist, DC, leakage and the Hann window

Reflect

Section titled “Reflect”
  • If you chained Median followed by EMA, what result would you expect, and what would you be trading off?
  • Is the highest noise-down % always the best outcome? Think of a job that needs a fast response.

Review questions

Answer on your own first, then open the answer.

  1. After filling the file, the Filtered line matches Raw exactly whatever filter is chosen. Which point is still empty? (Objective 1)

    1. return dsp.Median(window=MED_WINDOW)
    2. mag = (ax*ax + ay*ay + az*az) ** 0.5
    3. y = filt.update(x)
    4. raw_chart.value(clamp(x))
    Show answer

    Answer: C. y = filt.update(x)

    placeholder y = x ทำให้ค่าที่วาดบนเส้นล่างคือค่าดิบ สองกราฟต่างกันก็ต่อเมื่อค่าผ่าน filt.update จริง

  2. You write return dsp.Median(5) at point 1 and choose Median. What happens? (Objective 1)

    1. ได้หน้าต่าง 5 ค่าตามต้องการ
    2. โยน TypeError เพราะ window เป็นพารามิเตอร์แบบ keyword เท่านั้น
    3. ได้หน้าต่าง 6 ค่า
    4. ฟิลเตอร์กลายเป็น EMA
    Show answer

    Answer: B. โยน TypeError เพราะ window เป็นพารามิเตอร์แบบ keyword เท่านั้น

    พารามิเตอร์ของฟิลเตอร์ใน dsp เป็น keyword-only ต้องเขียน dsp.Median(window=5) การส่งแบบ positional จะถูกปฏิเสธทันที

  3. You tap the board hard once to make a spike. What do you expect on the Filtered line for EMA versus Median? (Objective 2)

    1. ทั้งสองตัดทิ้งได้เท่ากัน
    2. EMA เกลี่ยเป็นโหนกนุ่มที่ยังเห็น ส่วน Median โหวต spike ตกไปจนแทบไม่เห็น
    3. EMA ตัดทิ้งหมด ส่วน Median เกลี่ยเป็นโหนก
    4. ทั้งสองทำให้เส้นกลายเป็นศูนย์
    Show answer

    Answer: B. EMA เกลี่ยเป็นโหนกนุ่มที่ยังเห็น ส่วน Median โหวต spike ตกไปจนแทบไม่เห็น

    EMA ถ่วงค่าโดดเข้าไปในผลลัพธ์บางส่วน Median เลือกค่ากลางจึงไม่สนค่าโดดเดี่ยว นี่คือความต่างของ "เกลี่ย" กับ "โหวตตัด"

  4. raw_jit = 400 and filt_jit = 100. What is noise down %? (Objective 3)

    1. 25%
    2. 75%
    3. 100%
    4. 400%
    Show answer

    Answer: B. 75%

    red = (1 − 100/400) × 100 = 75% การกระตุกของเส้นกรองเหลือหนึ่งในสี่ของเส้นดิบ

Cite this lesson

If you teach from this lesson or reuse it in slides or documents, credit it with the text below. If you changed it, add (adapted) after the title.

"Hands-on: cleaning a live signal with a filter" from TESA Open Knowledge by the Thai Embedded Systems Association (TESA), https://github.com/tesaiot/tesa-qualification-program, licensed under CC BY-NC 4.0

Thai attribution: "ลงมือทำ: ฟิลเตอร์ทำสัญญาณให้สะอาดสด ๆ" จาก TESA Open Knowledge โดยสมาคมสมองกลฝังตัวไทย (Thai Embedded Systems Association: TESA) https://github.com/tesaiot/tesa-qualification-program สัญญาอนุญาต CC BY-NC 4.0

Lesson link: https://tesaiot.github.io/tesa-qualification-program/en/courses/edge-ai-developer/m04-analysis/l02-filters-lab/

Full guide: how to cite TESA

TESA Open Knowledge · © 2026 สมาคมสมองกลฝังตัวไทย (TESA) · CC BY-NC 4.0

Content is licensed CC BY-NC 4.0. Reuse it non-commercially and credit the Thai Embedded Systems Association (TESA) every time. · How to cite TESA