تجربة عملية: قارِن بين الصوت التناظري والرقمي
رقمِن نغمة، ثم أعِد بناءها
ما ستفعله
- ما سرعة التقاطها
- الهامش الذي يتركه صوت الأقراص
- تناظري يخرج — محوّل حقيقي
صفحات تفاعلية لهذا المقرر. كل رقم فيها يُحسَب في متصفّحك — فليست رسوماً مسجّلة، ولا تتّصل بأي خادم.
صفحات العروض بالإنجليزية في الوقت الحالي؛ أما الدرس المرتبط بكل عرض فهو بالعربية بالكامل.
رقمِن نغمة، ثم أعِد بناءها
خُذ عيّنات نغمة، وراقبها تنطوي
التفاف لحظة بلحظة
ضع قطبًا، وراقب الاستجابة تتحرّك
شاهِد نغمة تقع في خانة
اقرأ ما تكلّفه النافذة وما تشتريه
صمِّم مرشِّح FIR، واقرأ طوره الخطّي
قارِن FIR بـ IIR
اقرأ مقايضة الزمن والتردّد
انتشل إشارة من الضجيج
غيّر معدّل الأخذ، واحفظ الصوت
صمّم سلسلة واحدة، مرحلةً بعد مرحلة
Raise a tone past half the sampling rate and watch — and hear — the pitch come back down. The replica spectrum, the folding map, a real Butterworth anti-alias filter and the wagon-wheel view of the same arithmetic.
Build your own x and h by dragging the stems, then walk the lag while h flips, shifts and multiplies. Every product is written out term by term, the algebraic properties are recomputed on your sequences, and a computed room response lets you hear what a convolution is for.
Drag poles and zeros on the z-plane and watch four views move together: magnitude, phase, impulse response and the pole-zero map itself.
Where spectral leakage comes from: change the tone, the window and the zero-padding, and watch the DFT bins respond.
The window figures of merit, computed from the coefficients: equivalent noise bandwidth N·Σw²/(Σw)², coherent gain (Σw)/N, scalloping loss and peak sidelobe level, each shown beside the published Harris (1978) value. Put a tone (plus noise) through a windowed FFT, detect the peak, and read its frequency and its coherent-gain-corrected amplitude 2|X_peak|/Σw. Watch the analyzer's central trade: a narrow window resolves close tones but hides weak ones; a wide one sees the weak tone but merges the close pair.
Design the same specification as FIR and as IIR at once, and see the trade-off the lessons describe in words — order, phase linearity and stopband.
The time-frequency resolution trade-off, made physical: change the window length and overlap and watch what the spectrogram can and cannot resolve.
Bury a known template in seeded Gaussian noise, run the matched filter (a cross-correlation with the template), and watch the peak stand out where the raw record shows nothing. The processing gain 10*log10(N), the output SNR, and the ROC -- P_d against P_fa -- are all computed, with an optional Monte-Carlo overlay that converges onto the analytic curve.
Convert a signal from one sample rate to another by a rational factor L/M -- upsample by L, lowpass, downsample by M. The reduced L/M (48000 → 44100 = 147/160), the new sample count N' = round(N·L/M), and the anti-alias cutoff min(f_s, f_s')/2 are all computed, and zero-order-hold, linear and windowed-sinc interpolation are ranked by their reconstruction error (sinc < linear < ZOH). Downsample a high tone with the anti-alias filter off and watch it fold.