Uru:
- Ikike iji wuo na itinye nlele, Nyquist na aliasing echiche na nkwado AI
- Ikike ịsụgharị FFT, nzacha imewe na usoro windo n'ime koodu nhazi yana ngụkọ na AI
- Ikike iji nyochaa nsonaazụ nhazi mgbaàmà nke AI mepụtara site na iji akara ntụaka, njikwa otu na nha
Nhazi mgbaama bụ obi elektrọnik na injinia nkwukọrịta: nka nke iwere ikuku voltaji, ndekọ ụda, ọgụgụ ihe mmetụta, ma ọ bụ akara redio na iwepụta ozi dị n'ime ya, hichaa mkpọtụ ya na ịgbanwe ya. Na nkeji a, ị ga-ahụ ka esi eji AI mụta echiche nhazi mgbaàmà, wuo nzacha na mgbako mgbanwe, wee jiri koodu Python rụọ nyocha. Ma ịdọ aka ná ntị site na mmalite: obere njehie echiche na nhazi mgbaàmà (ugboro nlele na-ezighị ezi, windo windowing, interfering unit) ga-agbagọ kpamkpam na ndopuyo; Eserese ahụ ka dị mma, mana ọ na-ezighi ezi. Ọ bụ ya mere o ji dị mkpa iji nwalee nsonaazụ ọ bụla AI na-emepụta megide akara ntụaka ama ama na njikwa otu.
Sampling na Nyquist: ebe ọ na-amalite
Iji hazie mgbaàmà analọg (efegharị na-aga n'ihu) na kọmpụta, anyị na-atụ ya n'oge ụfọdụ wee gbanwee ya na ọnụọgụ; A na-akpọ nke a sample. Ọ bụ ugboro nlele (fs, unit Hz) ole sample anyị na-ewere kwa nkeji. Usoro Nyquist na-ekwu na iji nwetaghachi mgbaama nke ọma, oge nlele ga-abụrịrị opekata mpe ugboro abụọ kachasị na mgbama. Ọ bụrụ na emezughị ọnọdụ a, ụda dị elu na-amalite ịdị ka obere ugboro; A na-akpọ nke a aliasing na data ahụ mebiri emebi kpamkpam n'ihi na ịnweghị ike ịmata ọdịiche dị n'etiti ọnọdụ abụọ a.
AI na-akọwa echiche a nke ọma ma nwee ike iji ihe atụ gosipụta ya. Ma ị na-ekpebi mgbe ị na-ahọrọ ugboro nlele gị: gịnị bụ bandwidth nke mgbaàmà gị n'ezie, ị nwere nzacha analog nke na-egbochi ya, olee ọsọ ADC gị (Analog-Digital Converter: mgbawa nke na-agbanwe akara ngosi analog na nọmba) na-arụ ọrụ n'ezie. AI nwere ike “atụ aro” ugboro nlele, mana ntụnye a bụ ntule na-enweghị atụta ụdịdị dị iche iche nke mgbama gị.
Ndụmọdụ: Mgbe ị na-ajụ AI ajụjụ nlele, kọwaa ugboro kachasị elu na bandwit nke mgbaàmà gị n'onwe gị. "Ọ bụrụ na akụkụ kachasị elu nke mgbaàmà m bụ ~ 4 kHz, kedu ka m ga-esi họrọ fs na ebee ka m ga-etinye ihe nzacha mgbochi?" dị ka. Emela ka AI na-eche ụdịdị dị iche iche; tụọ.
FFT na nyocha ugboro
FFT (Fast Fourier Transform) bụ ngwa ngwa algọridim na-atụgharị akara site na axis oge gaa na axis ugboro. Yabụ "Kedu ugboro ole dị na mgbama a yana ike ole?" zaa ajụjụ a. AI na-eme nke ọma ka esi etinye FFT, kọwapụta nsonaazụ yana dee koodu Python (numpy/scipy). Mana enwere mmejọ ndị a na-ahụkarị na FFT, na koodu AI nwere ike ịgụnye ha n'ime nzuzo:
- Ugboro axis scaling: Axis ugboro nke mmepụta FFT dabere na ugboro nlele. Ọ bụrụ na abanyeghị fs na-ezighi ezi, a na-ewepụta ugboro niile na ezighi ezi, mana eserese ahụ na-egosi na ọ dị mma.
- Window: Mgbe enwetara mpempe mgbaama dị oke, nkwụsị na-apụta na nsọtụ ma na-ahụ mgbasa ozi pseudo (leakage spectral) na ụdịdị dị iche iche. Windows dị ka Hann na Hamming na-ebelata nke a. Ọ bụrụ na koodu AI na-amafe windo, nsonaazụ ya ga-eduhie eduhie.
- Nha nha nha: achọrọ ntule na mezie windo iji tọghata njupụta nke mmepụta FFT ka ọ bụrụ uru anụ ahụ n'ezie. Ọ bụrụ na ewepụrụ ya, njupụta ahụ agaghị ezighi ezi.
nyo imewe
Ihe nzacha bụ sekit ma ọ bụ algọridim na-ewepụta ugboro ugboro na-achọghị na mgbama. Enwere ụdị isi anọ: obere ngafe (na-agafe obere frequencies), ngafe dị elu, band-pass, na band-pass (notch). Enwere ezinụlọ abụọ buru ibu nke nzacha dijitalụ: FIR (nzaghachi mkpali zuru oke, kwụsiri ike na usoro ahịrị) na IIR (nzaghachi mkpali na-enweghị ngwụcha, ngbanwe dị elu na ngụkọ dị obere mana nkwụsi ike na ihe egwu oge).
AI na-akọwa usoro nhazi nke nzacha (ugboro nkwụsị, paspọd, nkwụsị nkwụsị, nhọrọ ịtụ) wee jiri scipy dee koodu na-emepụta ọnụọgụgụ. Mana oge nkwụsị nke nzacha ị chepụtara bụ nke a na-ahazi (n'ihe gbasara ugboro Nyquist), na ịbanye fs na-ezighi ezi ebe a bụ mmejọ a na-ahụkarị. Na mgbakwunye, nkwụsi ike na mgbagha oge nwere ike ịbụ nsogbu na nzacha IIR, na ịla azụ nwere ike bụrụ nsogbu na sistemu oge. Jide n'aka na iji eserese nzaghachi ugboro ugboro na akara ule amaara nyochaa nzacha AI mepụtara.
Obere ikpe atọ
Ikpe 1 - ịma jijiji na-atụgharị. Onye injinia na-enyocha ịma jijiji nke moto na 1 kHz wee hụ ọnụ ọgụgụ siri ike na FFT na 300 Hz. Ọnụ ọgụgụ kasị elu a, nke na-adabaghị na nhazi engine, na-agbagwoju anya. E nwere n'ezie a akụrụngwa na engine gburugburu 1300 Hz; 1 kHz nlele ezughị maka nke a (Nyquist 500 Hz) na 1300 Hz kpuchiri 300 Hz. Mgbe gị na AI kwurịtara "ma ọnụ ọgụgụ a dị adị ma ọ bụ na-ekewapụ" ma na-abawanye ugboro ugboro na 5 kHz, ọnụ ọgụgụ kasị elu na-apụta n'ebe kwesịrị ekwesị na 1300 Hz. Ihe ọmụmụ: n'elu ugwu a na-ajụ ajụjụ, ajụjụ mbụ na-egosi ntozu oke.
Ikpe 2 - mpio amafere. Onye na-arụ ọrụ na-arịọ maka koodu FFT site na AI wee mee ya. Ọ na-atụ anya otu ụda ule ugboro ugboro mana ọ na-ahụ ka agbasawanye n'ofe ụdịdị ahụ. Nsogbu bụ na koodu anaghị eme windowing ( leakage spectral ). Gwa AI "tinye windo Hann wee tinye mgbazi njupụta" na mmepụta ya laghachi na ọnụ ọgụgụ dị nkọ a tụrụ anya ya. Ihe nkuzi: Jiri akara mgbaama otu ugboro nwalee koodu FFT; Lelee mpio ahụ ma ọ bụrụ na mmiri dị.
Ikpe 3 - ezighi ezi fs. Onye injinia na-enyocha data sitere na onye na-edekọ ihe. Ngwaọrụ e dekọrọ na 48 kHz, mana 44.1 kHz ndabara etinyere na koodu AI. A na-atụgharị ugboro niile site na 8%; Ụda 1000 Hz na-egosi na 1088 Hz. A na-achọta mperi mgbe ị na-elele fs n'ezie nke akụrụngwa ihe atụ. Ihe nkuzi: oge nlele bụ otu ọnụọgụ kachasị dị egwu na agbụ nyocha; a ga-enyocharịrị site na isi mmalite.
Ndebiri ozugbo enwere ike iṅomi
FFT SETUP TEMPLATE"Dee koodu nyocha FFT na Python (numpy/scipy). Ntinye: samplingfrequency fs=[uru] Hz, eriri akara 'x'. Ka koodu ahụ mee ihe ndị a: (1) Tinye windo Hann, (2) Nweta FFT, (3) n'ụzọ ziri ezi na otu nha nha nha nhata n'akụkụ fs, (4) mpio nke ọ bụla n'ime nha nha nha nha n'akụkụ fs, (4). nke koodu "Biko kwupụta n'ahịrị nkọwa nke echiche ị na-eme ka m nwee ike nyochaa ya."
AKWỤKWỌ NKE NKWỤKWỌ NCHE AKWỤKWỌ "Oge mgbama m kachasi elu bụ ihe dịka [X] Hz. Kọwaa oke Nyquist, mkpa maka nzacha mgbochi, yana oke nchekwa bara uru mgbe ị na-ahọrọ ugboro nlele. Tụọ aro nso mana mesie ike na m ga-enyocha nke a site n'ịtụle ụdị mgbama ahụ n'ezie."
FILTER DESIGN TEMPLATE "Chepụta [obere / elu / band] gafere dijitalụ nzacha na scipy. Cutoff ugboro [X] Hz, samplingfrequency fs = [Y] Hz. Tụlee nhọrọ FIR na IIR (nkwụsi ike, oge, igbu oge, ibu mgbakọ).
AKWỤKWỌ NKE RESULT PRILIDATION "Nyochaa koodu nhazi akara n'okpuru yana ọkọlọtọ isi mmalite nke njehie dị jụụ: fs na-ezighị ezi, windo skipped, crosstalk unit (Hz / kHz, V / mV), enweghị normalization, otu / akụkụ abụọ.
Ngwa ngwa adịghị ike / Ike siri ike
EKWU IKE: "Chọta ugboro nke mgbama a."
IKE KWESỊRỊ: "Enwere m ihe nrịbama nke nwere ihe nlele nke 10 kHz. Dee koodu Python nke na-achọpụta ihe ndị na-emekarị ugboro ugboro na FFT; jiri Hann window na amplitude mmezi, mee ka axis ugboro dị ka fs. Na njedebe nke koodu ahụ, gbakwunye ngọngọ nkwenye na mgbama ule synthetic nke 1 kHz fs ma hụ ma ọ bụrụ na scaling. "
Ngwa ngwa na-adịghị ike na-ewepụta nzaghachi na-enweghị ihe ọ bụla wee were fs; Ọ na-eweghachi paramita ngwa ngwa siri ike ma kpalie maka ngọngọ nyocha nke onwe.
Tebụl nke nkeji na echiche na nhazi mgbaàmà
echiche
njehie ugboro ugboro
nkwenye
Oge nlele (fs)
Ndi an-kpọ kama FS nke ngwaọrụ n'ezie
Nkwenye site na ntọala ndekọ
axis ugboro ugboro
Ọ bụrụ na fs ezighi ezi, axis niile na-agbanwe
Ụda ule ugboro ama ama
Ime windo
Ntupu spektral mgbe awụsara ya
Obosara kacha elu na mgbama monotone
Njupụta
Nhazi/mmezi na-efu
Tụnyere akara njupụta amaara ama
nyo ịkpụ
normalize njehie dị ka fs
Eserese nzaghachi ugboro ugboro
Akpachara anya: Naanị n'ihi na eserese FFT na-adị ire ụtọ yana ikwere apụtaghị na ọ bụ eziokwu. Ọ bụrụ na agbanyere axis ugboro ugboro na-ezighi ezi, ọnụ ọgụgụ kacha elu niile nọ n'ebe na-ezighi ezi mana eserese ahụ na-adị ka nke zuru oke. Jiri ntụaka amara ama ama.
Mmejọ ndị nkịtị
- Anaghị enyocha ugboro nlele site na isi mmalite. fs na-ezighi ezi na-agbanwe ugboro niile.
- Na-emebi Nyquist na achọpụtaghị aliasing. N'elu ugwu ndị a na-enyo enyo, buru ụzọ jụọ maka ntozu nlele.
- Mwụpụ windo na FFT. Mgbapụta Spectral dị ka ezigbo ọnụ ọgụgụ.
- Na-echefu njupụta normalization. Njupụta na-anọgide na-enweghị atụ ruo uru anụ ahụ.
- Na-emezigharị ugboro ole nzacha nzacha na-ezighi ezi ka ọ bụrụ fs. Nzacha na-egbutu ebe dị iche karịa ka a tụrụ anya ya.
Na nchịkọta
N'ime nkeji a, ị dobere AI dị ka ihe enyemaka dị ike n'ịmụ echiche nhazi akara, ide FFT na koodu nzacha, na ịmepụta nyocha. Mana na nhazi akara ngosi, nsonaazụ ya nwere ike gbagọọ nwayọ: ugboro ole nlele na-ezighi ezi, ịfefe windo, olu na-egbochi ya, ma ọ bụ enweghị nhazi na-eme ka eserese ahụ maa mma mana ezighi ezi. Ya mere nwalee koodu ọ bụla AI na-emepụta megide akara ngosi ama ama, nyochaa oge nlele site na isi mmalite, wee nwalee nsonaazụ ọ bụla maka nkeji na ịdị mma anụ ahụ. Lelee echiche nke Nyquist na aliasing na mmalite nke nyocha ọ bụla.
Ọrụ ngwa
Mepụta mgbama ule sịntetik na Python: nchikota nke ugboro abụọ amaara (dịka 1 kHz na 3 kHz), fs=10 kHz. Rịọ koodu nyocha site na AI na ndebiri “FFT setup” wee chọpụta na ọnụ ọgụgụ ndị ahụ na-apụta na ugboro ole ziri ezi. Wee kpachaara anya banye fs na-ezighi ezi (dịka 8 kHz) wee hụ ka ọnụ ọgụgụ ahụ si agbanwe. N'ikpeazụ, wepụ windo ahụ ma hụ ntapu nke spectral. Dee nsonaazụ nnwale ọ bụla n'otu ahịrịokwu.
ndetu
- [] Achọpụtara m ugboro nlele (fs) site na ntọala n'ezie nke ihe ndekọ.
- [] Elelere m akara Nyquist wee nyochaa ihe egwu dị n'ịkpọ aha.
- [] Achọpụtala m na a na-emejuputa ngbanwe window na njupụta na koodu FFT.
- [ ] Eji m nrịbama nrụtụ aka nke ugboro ama ama.
- [] Eji m nzaghachi anya gosi na a na-ahazi ugboro ole nzacha nzacha nke ọma ka ọ bụrụ fs.
- [] Emere m nkeji niile (Hz/kHz, V/mV) na-agbanwe agbanwe.